program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}})]
{
    func main<ios16>(tensor<fp16, [2, 1024, 1, 77]> encoder_hidden_states, tensor<fp16, [2, 4, 64, 64]> sample, tensor<fp16, [2]> timestep) {
            tensor<int32, []> var_25 = const()[name = tensor<string, []>("op_25"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> var_42_axes_0 = const()[name = tensor<string, []>("op_42_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1]> var_42_cast = expand_dims(axes = var_42_axes_0, x = timestep)[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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(448))), lut = tensor<fp16, [4]>([-0x1.7d4p-4, -0x1.81cp-8, 0x1.168p-8, 0x1.0dp-4]), 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, []>(102912)))];
            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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105536))), lut = tensor<fp16, [4]>([-0x1.2e4p-4, -0x1.438p-10, 0x1.3e8p-10, 0x1.514p-4]), 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, []>(515200)))];
            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, [2880]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517824))), lut = tensor<fp16, [4]>([-0x1.f78p-4, -0x1.ee4p-6, 0x1.5d8p-6, 0x1.b1cp-4]), 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, []>(520768)))];
            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_115 = const()[name = tensor<string, []>("op_115"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_120 = const()[name = tensor<string, []>("op_120"), 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, []>(521472)))];
            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, []>(522176)))];
            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, []>(522880)))];
            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, []>(523584)))];
            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_142 = const()[name = tensor<string, []>("op_142"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_144 = const()[name = tensor<string, []>("op_144"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(524288))), lut = tensor<fp16, [4]>([-0x1.8dcp-4, -0x1.fa8p-6, 0x1.208p-6, 0x1.3dp-4]), 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, []>(754752)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_1_cast = conv(bias = down_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_144, groups = var_120, pad = hidden_states_1_pad_0, pad_type = hidden_states_1_pad_type_0, strides = var_142, 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_150 = const()[name = tensor<string, []>("op_150"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_152 = const()[name = tensor<string, []>("op_152"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(755456))), lut = tensor<fp16, [4]>([-0x1.154p-1, -0x1.414p-4, -0x1.8ap-16, 0x1.5p-4]), 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, []>(857920)))];
            tensor<fp16, [2, 320, 1, 1]> temb_1_cast = conv(bias = down_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_152, groups = var_120, pad = temb_1_pad_0, pad_type = temb_1_pad_type_0, strides = var_150, 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, []>(858624)))];
            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, []>(859328)))];
            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_162 = const()[name = tensor<string, []>("op_162"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_164 = const()[name = tensor<string, []>("op_164"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(860032))), lut = tensor<fp16, [4]>([-0x1.4fcp-4, -0x1.504p-6, 0x1.4f8p-6, 0x1.4dp-4]), 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, []>(1090496)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_3_cast = conv(bias = down_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_164, groups = var_120, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_162, 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, []>(1091200)))];
            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, []>(1091904)))];
            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_184 = const()[name = tensor<string, []>("op_184"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_186 = const()[name = tensor<string, []>("op_186"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1092608))), lut = tensor<fp16, [4]>([-0x1.7c4p-4, -0x1.cccp-6, 0x1.afcp-6, 0x1.734p-4]), 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, []>(1118272)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_7_cast = conv(bias = down_blocks_0_attentions_0_proj_in_bias_to_fp16, dilations = var_186, groups = var_120, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = var_184, 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_191 = const()[name = tensor<string, []>("op_191"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_1_cast = reshape(shape = var_191, x = hidden_states_7_cast)[name = tensor<string, []>("inputs_1_cast")];
            tensor<int32, [1]> var_201 = const()[name = tensor<string, []>("op_201"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_1_cast = reduce_mean(axes = var_201, keep_dims = var_115, 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_205 = const()[name = tensor<string, []>("op_205"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_206_cast = reduce_mean(axes = var_205, keep_dims = var_115, x = zero_mean_sq_1_cast)[name = tensor<string, []>("op_206_cast")];
            tensor<fp16, []> var_207_to_fp16 = const()[name = tensor<string, []>("op_207_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_208_cast = add(x = var_206_cast, y = var_207_to_fp16)[name = tensor<string, []>("op_208_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_208_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_212_to_fp16 = const()[name = tensor<string, []>("op_212_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1118976)))];
            tensor<fp16, [2, 320, 1, 4096]> var_213_cast = add(x = out_1_cast, y = var_212_to_fp16)[name = tensor<string, []>("op_213_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, []>(1119680)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_9_cast = mul(x = var_213_cast, y = var_215_to_fp16)[name = tensor<string, []>("hidden_states_9_cast")];
            tensor<int32, [2]> var_222 = const()[name = tensor<string, []>("op_222"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_224 = const()[name = tensor<string, []>("op_224"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120384))), lut = tensor<fp16, [4]>([-0x1.588p-3, -0x1.8ep-5, 0x1.894p-5, 0x1.574p-3]), 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_224, groups = var_120, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = var_222, 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_228 = const()[name = tensor<string, []>("op_228"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_230 = const()[name = tensor<string, []>("op_230"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1146048))), lut = tensor<fp16, [4]>([-0x1.3c8p-3, -0x1.654p-5, 0x1.70cp-5, 0x1.42cp-3]), 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_230, groups = var_120, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = var_228, 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_234 = const()[name = tensor<string, []>("op_234"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_236 = const()[name = tensor<string, []>("op_236"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171712))), lut = tensor<fp16, [4]>([-0x1.63cp-4, -0x1.8f4p-6, 0x1.918p-6, 0x1.628p-4]), 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_236, groups = var_120, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = var_234, 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_240_begin_0 = const()[name = tensor<string, []>("op_240_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_240_end_0 = const()[name = tensor<string, []>("op_240_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_240_end_mask_0 = const()[name = tensor<string, []>("op_240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_240_cast = slice_by_index(begin = var_240_begin_0, end = var_240_end_0, end_mask = var_240_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_240_cast")];
            tensor<int32, [4]> var_244_begin_0 = const()[name = tensor<string, []>("op_244_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_244_end_0 = const()[name = tensor<string, []>("op_244_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_244_end_mask_0 = const()[name = tensor<string, []>("op_244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_244_cast = slice_by_index(begin = var_244_begin_0, end = var_244_end_0, end_mask = var_244_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_244_cast")];
            tensor<int32, [4]> var_248_begin_0 = const()[name = tensor<string, []>("op_248_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_248_end_0 = const()[name = tensor<string, []>("op_248_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_248_end_mask_0 = const()[name = tensor<string, []>("op_248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_248_cast = slice_by_index(begin = var_248_begin_0, end = var_248_end_0, end_mask = var_248_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_248_cast")];
            tensor<int32, [4]> var_252_begin_0 = const()[name = tensor<string, []>("op_252_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_252_end_0 = const()[name = tensor<string, []>("op_252_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_252_end_mask_0 = const()[name = tensor<string, []>("op_252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_252_cast = slice_by_index(begin = var_252_begin_0, end = var_252_end_0, end_mask = var_252_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_252_cast")];
            tensor<int32, [4]> var_256_begin_0 = const()[name = tensor<string, []>("op_256_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_256_end_0 = const()[name = tensor<string, []>("op_256_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_256_end_mask_0 = const()[name = tensor<string, []>("op_256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_256_cast = slice_by_index(begin = var_256_begin_0, end = var_256_end_0, end_mask = var_256_end_mask_0, x = q_1_cast)[name = tensor<string, []>("op_256_cast")];
            tensor<int32, [4]> var_259_begin_0 = const()[name = tensor<string, []>("op_259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_259_end_0 = const()[name = tensor<string, []>("op_259_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_259_end_mask_0 = const()[name = tensor<string, []>("op_259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_259_cast = slice_by_index(begin = var_259_begin_0, end = var_259_end_0, end_mask = var_259_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_259_cast")];
            tensor<int32, [4]> var_260_begin_0 = const()[name = tensor<string, []>("op_260_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_260_end_0 = const()[name = tensor<string, []>("op_260_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_260_end_mask_0 = const()[name = tensor<string, []>("op_260_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_260_cast = slice_by_index(begin = var_260_begin_0, end = var_260_end_0, end_mask = var_260_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_260_cast")];
            tensor<int32, [4]> var_261_begin_0 = const()[name = tensor<string, []>("op_261_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_261_end_0 = const()[name = tensor<string, []>("op_261_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_261_end_mask_0 = const()[name = tensor<string, []>("op_261_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_261_cast = slice_by_index(begin = var_261_begin_0, end = var_261_end_0, end_mask = var_261_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_261_cast")];
            tensor<int32, [4]> var_262_begin_0 = const()[name = tensor<string, []>("op_262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_262_end_0 = const()[name = tensor<string, []>("op_262_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_262_end_mask_0 = const()[name = tensor<string, []>("op_262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_262_cast = slice_by_index(begin = var_262_begin_0, end = var_262_end_0, end_mask = var_262_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_262_cast")];
            tensor<int32, [4]> var_263_begin_0 = const()[name = tensor<string, []>("op_263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_263_end_0 = const()[name = tensor<string, []>("op_263_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_263_end_mask_0 = const()[name = tensor<string, []>("op_263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_263_cast = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_263_cast")];
            tensor<int32, [4]> var_264_begin_0 = const()[name = tensor<string, []>("op_264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_264_end_0 = const()[name = tensor<string, []>("op_264_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_264_end_mask_0 = const()[name = tensor<string, []>("op_264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_264_cast = slice_by_index(begin = var_264_begin_0, end = var_264_end_0, end_mask = var_264_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_264_cast")];
            tensor<int32, [4]> var_265_begin_0 = const()[name = tensor<string, []>("op_265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_265_end_0 = const()[name = tensor<string, []>("op_265_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_265_end_mask_0 = const()[name = tensor<string, []>("op_265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_265_cast = slice_by_index(begin = var_265_begin_0, end = var_265_end_0, end_mask = var_265_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_265_cast")];
            tensor<int32, [4]> var_266_begin_0 = const()[name = tensor<string, []>("op_266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_266_end_0 = const()[name = tensor<string, []>("op_266_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_266_end_mask_0 = const()[name = tensor<string, []>("op_266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_266_cast = slice_by_index(begin = var_266_begin_0, end = var_266_end_0, end_mask = var_266_end_mask_0, x = var_240_cast)[name = tensor<string, []>("op_266_cast")];
            tensor<int32, [4]> var_267_begin_0 = const()[name = tensor<string, []>("op_267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_267_end_0 = const()[name = tensor<string, []>("op_267_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_267_end_mask_0 = const()[name = tensor<string, []>("op_267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_267_cast = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_267_cast")];
            tensor<int32, [4]> var_268_begin_0 = const()[name = tensor<string, []>("op_268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_268_end_0 = const()[name = tensor<string, []>("op_268_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_268_end_mask_0 = const()[name = tensor<string, []>("op_268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_268_cast = slice_by_index(begin = var_268_begin_0, end = var_268_end_0, end_mask = var_268_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_268_cast")];
            tensor<int32, [4]> var_269_begin_0 = const()[name = tensor<string, []>("op_269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_269_end_0 = const()[name = tensor<string, []>("op_269_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_269_end_mask_0 = const()[name = tensor<string, []>("op_269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_269_cast = slice_by_index(begin = var_269_begin_0, end = var_269_end_0, end_mask = var_269_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_269_cast")];
            tensor<int32, [4]> var_270_begin_0 = const()[name = tensor<string, []>("op_270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_270_end_0 = const()[name = tensor<string, []>("op_270_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_270_end_mask_0 = const()[name = tensor<string, []>("op_270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_270_cast = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_270_cast")];
            tensor<int32, [4]> var_271_begin_0 = const()[name = tensor<string, []>("op_271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_271_end_0 = const()[name = tensor<string, []>("op_271_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_271_end_mask_0 = const()[name = tensor<string, []>("op_271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_271_cast = slice_by_index(begin = var_271_begin_0, end = var_271_end_0, end_mask = var_271_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_271_cast")];
            tensor<int32, [4]> var_272_begin_0 = const()[name = tensor<string, []>("op_272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_272_end_0 = const()[name = tensor<string, []>("op_272_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_272_end_mask_0 = const()[name = tensor<string, []>("op_272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_272_cast = slice_by_index(begin = var_272_begin_0, end = var_272_end_0, end_mask = var_272_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_272_cast")];
            tensor<int32, [4]> var_273_begin_0 = const()[name = tensor<string, []>("op_273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_273_end_0 = const()[name = tensor<string, []>("op_273_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_273_end_mask_0 = const()[name = tensor<string, []>("op_273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_273_cast = slice_by_index(begin = var_273_begin_0, end = var_273_end_0, end_mask = var_273_end_mask_0, x = var_244_cast)[name = tensor<string, []>("op_273_cast")];
            tensor<int32, [4]> var_274_begin_0 = const()[name = tensor<string, []>("op_274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_274_end_0 = const()[name = tensor<string, []>("op_274_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            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, 64, 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_244_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, 0])];
            tensor<int32, [4]> var_275_end_0 = const()[name = tensor<string, []>("op_275_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_248_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, 512])];
            tensor<int32, [4]> var_276_end_0 = const()[name = tensor<string, []>("op_276_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_248_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, 1024])];
            tensor<int32, [4]> var_277_end_0 = const()[name = tensor<string, []>("op_277_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            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, 64, 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_248_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, 1536])];
            tensor<int32, [4]> var_278_end_0 = const()[name = tensor<string, []>("op_278_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            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, 64, 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_248_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, 2048])];
            tensor<int32, [4]> var_279_end_0 = const()[name = tensor<string, []>("op_279_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            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, 64, 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_248_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, 2560])];
            tensor<int32, [4]> var_280_end_0 = const()[name = tensor<string, []>("op_280_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            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, 64, 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_248_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, 3072])];
            tensor<int32, [4]> var_281_end_0 = const()[name = tensor<string, []>("op_281_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            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, 64, 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_248_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, 3584])];
            tensor<int32, [4]> var_282_end_0 = const()[name = tensor<string, []>("op_282_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            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, 64, 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_248_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, 0])];
            tensor<int32, [4]> var_283_end_0 = const()[name = tensor<string, []>("op_283_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_252_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, 512])];
            tensor<int32, [4]> var_284_end_0 = const()[name = tensor<string, []>("op_284_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_252_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, 1024])];
            tensor<int32, [4]> var_285_end_0 = const()[name = tensor<string, []>("op_285_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            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, 64, 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_252_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, 1536])];
            tensor<int32, [4]> var_286_end_0 = const()[name = tensor<string, []>("op_286_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            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, 64, 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_252_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, 2048])];
            tensor<int32, [4]> var_287_end_0 = const()[name = tensor<string, []>("op_287_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            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, 64, 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_252_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, 2560])];
            tensor<int32, [4]> var_288_end_0 = const()[name = tensor<string, []>("op_288_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            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, 64, 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_252_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, 3072])];
            tensor<int32, [4]> var_289_end_0 = const()[name = tensor<string, []>("op_289_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            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, 64, 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_252_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, 3584])];
            tensor<int32, [4]> var_290_end_0 = const()[name = tensor<string, []>("op_290_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            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, 64, 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_252_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, 0])];
            tensor<int32, [4]> var_291_end_0 = const()[name = tensor<string, []>("op_291_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_256_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, 512])];
            tensor<int32, [4]> var_292_end_0 = const()[name = tensor<string, []>("op_292_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_256_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, 1024])];
            tensor<int32, [4]> var_293_end_0 = const()[name = tensor<string, []>("op_293_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            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, 64, 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_256_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, 1536])];
            tensor<int32, [4]> var_294_end_0 = const()[name = tensor<string, []>("op_294_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            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, 64, 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_256_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, 2048])];
            tensor<int32, [4]> var_295_end_0 = const()[name = tensor<string, []>("op_295_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            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, 64, 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_256_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, 2560])];
            tensor<int32, [4]> var_296_end_0 = const()[name = tensor<string, []>("op_296_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            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, 64, 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_256_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, 3072])];
            tensor<int32, [4]> var_297_end_0 = const()[name = tensor<string, []>("op_297_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            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, 64, 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_256_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, 3584])];
            tensor<int32, [4]> var_298_end_0 = const()[name = tensor<string, []>("op_298_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            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, 64, 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_256_cast)[name = tensor<string, []>("op_298_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_303_begin_0 = const()[name = tensor<string, []>("op_303_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_303_end_0 = const()[name = tensor<string, []>("op_303_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            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, 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, 64]> var_303_cast = slice_by_index(begin = var_303_begin_0, end = var_303_end_0, end_mask = var_303_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_303_cast")];
            tensor<int32, [4]> var_307_begin_0 = const()[name = tensor<string, []>("op_307_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_307_end_0 = const()[name = tensor<string, []>("op_307_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            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, 4096, 1, 64]> var_307_cast = slice_by_index(begin = var_307_begin_0, end = var_307_end_0, end_mask = var_307_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_307_cast")];
            tensor<int32, [4]> var_311_begin_0 = const()[name = tensor<string, []>("op_311_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_311_end_0 = const()[name = tensor<string, []>("op_311_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            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, 4096, 1, 64]> var_311_cast = slice_by_index(begin = var_311_begin_0, end = var_311_end_0, end_mask = var_311_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_311_cast")];
            tensor<int32, [4]> var_315_begin_0 = const()[name = tensor<string, []>("op_315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_315_end_0 = const()[name = tensor<string, []>("op_315_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            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, 4096, 1, 64]> var_315_cast = slice_by_index(begin = var_315_begin_0, end = var_315_end_0, end_mask = var_315_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_315_cast")];
            tensor<int32, [4]> var_319_begin_0 = const()[name = tensor<string, []>("op_319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_319_end_0 = const()[name = tensor<string, []>("op_319_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            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, 4096, 1, 64]> var_319_cast = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = transpose_31)[name = tensor<string, []>("op_319_cast")];
            tensor<int32, [4]> var_321_begin_0 = const()[name = tensor<string, []>("op_321_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_321_end_0 = const()[name = tensor<string, []>("op_321_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_321_end_mask_0 = const()[name = tensor<string, []>("op_321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_321_cast = slice_by_index(begin = var_321_begin_0, end = var_321_end_0, end_mask = var_321_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_321_cast")];
            tensor<int32, [4]> var_325_begin_0 = const()[name = tensor<string, []>("op_325_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_325_end_0 = const()[name = tensor<string, []>("op_325_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_325_end_mask_0 = const()[name = tensor<string, []>("op_325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_325_cast = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_325_cast")];
            tensor<int32, [4]> var_329_begin_0 = const()[name = tensor<string, []>("op_329_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_329_end_0 = const()[name = tensor<string, []>("op_329_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_329_end_mask_0 = const()[name = tensor<string, []>("op_329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_329_cast = slice_by_index(begin = var_329_begin_0, end = var_329_end_0, end_mask = var_329_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_329_cast")];
            tensor<int32, [4]> var_333_begin_0 = const()[name = tensor<string, []>("op_333_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_333_end_0 = const()[name = tensor<string, []>("op_333_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_333_end_mask_0 = const()[name = tensor<string, []>("op_333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_333_cast = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_333_cast")];
            tensor<int32, [4]> var_337_begin_0 = const()[name = tensor<string, []>("op_337_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_337_end_0 = const()[name = tensor<string, []>("op_337_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_337_end_mask_0 = const()[name = tensor<string, []>("op_337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_337_cast = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = v_1_cast)[name = tensor<string, []>("op_337_cast")];
            tensor<string, []> var_341_equation_0 = const()[name = tensor<string, []>("op_341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_341_cast = einsum(equation = var_341_equation_0, values = (var_303_cast, var_259_cast))[name = tensor<string, []>("op_341_cast")];
            tensor<fp16, []> var_342_to_fp16 = const()[name = tensor<string, []>("op_342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1_cast = mul(x = var_341_cast, y = var_342_to_fp16)[name = tensor<string, []>("aw_chunk_1_cast")];
            tensor<string, []> var_345_equation_0 = const()[name = tensor<string, []>("op_345_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_345_cast = einsum(equation = var_345_equation_0, values = (var_303_cast, var_260_cast))[name = tensor<string, []>("op_345_cast")];
            tensor<fp16, []> var_346_to_fp16 = const()[name = tensor<string, []>("op_346_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_3_cast = mul(x = var_345_cast, y = var_346_to_fp16)[name = tensor<string, []>("aw_chunk_3_cast")];
            tensor<string, []> var_349_equation_0 = const()[name = tensor<string, []>("op_349_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_349_cast = einsum(equation = var_349_equation_0, values = (var_303_cast, var_261_cast))[name = tensor<string, []>("op_349_cast")];
            tensor<fp16, []> var_350_to_fp16 = const()[name = tensor<string, []>("op_350_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_5_cast = mul(x = var_349_cast, y = var_350_to_fp16)[name = tensor<string, []>("aw_chunk_5_cast")];
            tensor<string, []> var_353_equation_0 = const()[name = tensor<string, []>("op_353_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_353_cast = einsum(equation = var_353_equation_0, values = (var_303_cast, var_262_cast))[name = tensor<string, []>("op_353_cast")];
            tensor<fp16, []> var_354_to_fp16 = const()[name = tensor<string, []>("op_354_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_7_cast = mul(x = var_353_cast, y = var_354_to_fp16)[name = tensor<string, []>("aw_chunk_7_cast")];
            tensor<string, []> var_357_equation_0 = const()[name = tensor<string, []>("op_357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_357_cast = einsum(equation = var_357_equation_0, values = (var_303_cast, var_263_cast))[name = tensor<string, []>("op_357_cast")];
            tensor<fp16, []> var_358_to_fp16 = const()[name = tensor<string, []>("op_358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_9_cast = mul(x = var_357_cast, y = var_358_to_fp16)[name = tensor<string, []>("aw_chunk_9_cast")];
            tensor<string, []> var_361_equation_0 = const()[name = tensor<string, []>("op_361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_361_cast = einsum(equation = var_361_equation_0, values = (var_303_cast, var_264_cast))[name = tensor<string, []>("op_361_cast")];
            tensor<fp16, []> var_362_to_fp16 = const()[name = tensor<string, []>("op_362_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_11_cast = mul(x = var_361_cast, y = var_362_to_fp16)[name = tensor<string, []>("aw_chunk_11_cast")];
            tensor<string, []> var_365_equation_0 = const()[name = tensor<string, []>("op_365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_365_cast = einsum(equation = var_365_equation_0, values = (var_303_cast, var_265_cast))[name = tensor<string, []>("op_365_cast")];
            tensor<fp16, []> var_366_to_fp16 = const()[name = tensor<string, []>("op_366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_13_cast = mul(x = var_365_cast, y = var_366_to_fp16)[name = tensor<string, []>("aw_chunk_13_cast")];
            tensor<string, []> var_369_equation_0 = const()[name = tensor<string, []>("op_369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_369_cast = einsum(equation = var_369_equation_0, values = (var_303_cast, var_266_cast))[name = tensor<string, []>("op_369_cast")];
            tensor<fp16, []> var_370_to_fp16 = const()[name = tensor<string, []>("op_370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_15_cast = mul(x = var_369_cast, y = var_370_to_fp16)[name = tensor<string, []>("aw_chunk_15_cast")];
            tensor<string, []> var_373_equation_0 = const()[name = tensor<string, []>("op_373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_373_cast = einsum(equation = var_373_equation_0, values = (var_307_cast, var_267_cast))[name = tensor<string, []>("op_373_cast")];
            tensor<fp16, []> var_374_to_fp16 = const()[name = tensor<string, []>("op_374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_17_cast = mul(x = var_373_cast, y = var_374_to_fp16)[name = tensor<string, []>("aw_chunk_17_cast")];
            tensor<string, []> var_377_equation_0 = const()[name = tensor<string, []>("op_377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_377_cast = einsum(equation = var_377_equation_0, values = (var_307_cast, var_268_cast))[name = tensor<string, []>("op_377_cast")];
            tensor<fp16, []> var_378_to_fp16 = const()[name = tensor<string, []>("op_378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_19_cast = mul(x = var_377_cast, y = var_378_to_fp16)[name = tensor<string, []>("aw_chunk_19_cast")];
            tensor<string, []> var_381_equation_0 = const()[name = tensor<string, []>("op_381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_381_cast = einsum(equation = var_381_equation_0, values = (var_307_cast, var_269_cast))[name = tensor<string, []>("op_381_cast")];
            tensor<fp16, []> var_382_to_fp16 = const()[name = tensor<string, []>("op_382_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_21_cast = mul(x = var_381_cast, y = var_382_to_fp16)[name = tensor<string, []>("aw_chunk_21_cast")];
            tensor<string, []> var_385_equation_0 = const()[name = tensor<string, []>("op_385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_385_cast = einsum(equation = var_385_equation_0, values = (var_307_cast, var_270_cast))[name = tensor<string, []>("op_385_cast")];
            tensor<fp16, []> var_386_to_fp16 = const()[name = tensor<string, []>("op_386_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_23_cast = mul(x = var_385_cast, y = var_386_to_fp16)[name = tensor<string, []>("aw_chunk_23_cast")];
            tensor<string, []> var_389_equation_0 = const()[name = tensor<string, []>("op_389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_389_cast = einsum(equation = var_389_equation_0, values = (var_307_cast, var_271_cast))[name = tensor<string, []>("op_389_cast")];
            tensor<fp16, []> var_390_to_fp16 = const()[name = tensor<string, []>("op_390_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_25_cast = mul(x = var_389_cast, y = var_390_to_fp16)[name = tensor<string, []>("aw_chunk_25_cast")];
            tensor<string, []> var_393_equation_0 = const()[name = tensor<string, []>("op_393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_393_cast = einsum(equation = var_393_equation_0, values = (var_307_cast, var_272_cast))[name = tensor<string, []>("op_393_cast")];
            tensor<fp16, []> var_394_to_fp16 = const()[name = tensor<string, []>("op_394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_27_cast = mul(x = var_393_cast, y = var_394_to_fp16)[name = tensor<string, []>("aw_chunk_27_cast")];
            tensor<string, []> var_397_equation_0 = const()[name = tensor<string, []>("op_397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_397_cast = einsum(equation = var_397_equation_0, values = (var_307_cast, var_273_cast))[name = tensor<string, []>("op_397_cast")];
            tensor<fp16, []> var_398_to_fp16 = const()[name = tensor<string, []>("op_398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_29_cast = mul(x = var_397_cast, y = var_398_to_fp16)[name = tensor<string, []>("aw_chunk_29_cast")];
            tensor<string, []> var_401_equation_0 = const()[name = tensor<string, []>("op_401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_401_cast = einsum(equation = var_401_equation_0, values = (var_307_cast, var_274_cast))[name = tensor<string, []>("op_401_cast")];
            tensor<fp16, []> var_402_to_fp16 = const()[name = tensor<string, []>("op_402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_31_cast = mul(x = var_401_cast, y = var_402_to_fp16)[name = tensor<string, []>("aw_chunk_31_cast")];
            tensor<string, []> var_405_equation_0 = const()[name = tensor<string, []>("op_405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_405_cast = einsum(equation = var_405_equation_0, values = (var_311_cast, var_275_cast))[name = tensor<string, []>("op_405_cast")];
            tensor<fp16, []> var_406_to_fp16 = const()[name = tensor<string, []>("op_406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_33_cast = mul(x = var_405_cast, y = var_406_to_fp16)[name = tensor<string, []>("aw_chunk_33_cast")];
            tensor<string, []> var_409_equation_0 = const()[name = tensor<string, []>("op_409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_409_cast = einsum(equation = var_409_equation_0, values = (var_311_cast, var_276_cast))[name = tensor<string, []>("op_409_cast")];
            tensor<fp16, []> var_410_to_fp16 = const()[name = tensor<string, []>("op_410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_35_cast = mul(x = var_409_cast, y = var_410_to_fp16)[name = tensor<string, []>("aw_chunk_35_cast")];
            tensor<string, []> var_413_equation_0 = const()[name = tensor<string, []>("op_413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_413_cast = einsum(equation = var_413_equation_0, values = (var_311_cast, var_277_cast))[name = tensor<string, []>("op_413_cast")];
            tensor<fp16, []> var_414_to_fp16 = const()[name = tensor<string, []>("op_414_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_37_cast = mul(x = var_413_cast, y = var_414_to_fp16)[name = tensor<string, []>("aw_chunk_37_cast")];
            tensor<string, []> var_417_equation_0 = const()[name = tensor<string, []>("op_417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_417_cast = einsum(equation = var_417_equation_0, values = (var_311_cast, var_278_cast))[name = tensor<string, []>("op_417_cast")];
            tensor<fp16, []> var_418_to_fp16 = const()[name = tensor<string, []>("op_418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_39_cast = mul(x = var_417_cast, y = var_418_to_fp16)[name = tensor<string, []>("aw_chunk_39_cast")];
            tensor<string, []> var_421_equation_0 = const()[name = tensor<string, []>("op_421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_421_cast = einsum(equation = var_421_equation_0, values = (var_311_cast, var_279_cast))[name = tensor<string, []>("op_421_cast")];
            tensor<fp16, []> var_422_to_fp16 = const()[name = tensor<string, []>("op_422_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_41_cast = mul(x = var_421_cast, y = var_422_to_fp16)[name = tensor<string, []>("aw_chunk_41_cast")];
            tensor<string, []> var_425_equation_0 = const()[name = tensor<string, []>("op_425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_425_cast = einsum(equation = var_425_equation_0, values = (var_311_cast, var_280_cast))[name = tensor<string, []>("op_425_cast")];
            tensor<fp16, []> var_426_to_fp16 = const()[name = tensor<string, []>("op_426_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_43_cast = mul(x = var_425_cast, y = var_426_to_fp16)[name = tensor<string, []>("aw_chunk_43_cast")];
            tensor<string, []> var_429_equation_0 = const()[name = tensor<string, []>("op_429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_429_cast = einsum(equation = var_429_equation_0, values = (var_311_cast, var_281_cast))[name = tensor<string, []>("op_429_cast")];
            tensor<fp16, []> var_430_to_fp16 = const()[name = tensor<string, []>("op_430_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_45_cast = mul(x = var_429_cast, y = var_430_to_fp16)[name = tensor<string, []>("aw_chunk_45_cast")];
            tensor<string, []> var_433_equation_0 = const()[name = tensor<string, []>("op_433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_433_cast = einsum(equation = var_433_equation_0, values = (var_311_cast, var_282_cast))[name = tensor<string, []>("op_433_cast")];
            tensor<fp16, []> var_434_to_fp16 = const()[name = tensor<string, []>("op_434_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_47_cast = mul(x = var_433_cast, y = var_434_to_fp16)[name = tensor<string, []>("aw_chunk_47_cast")];
            tensor<string, []> var_437_equation_0 = const()[name = tensor<string, []>("op_437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_437_cast = einsum(equation = var_437_equation_0, values = (var_315_cast, var_283_cast))[name = tensor<string, []>("op_437_cast")];
            tensor<fp16, []> var_438_to_fp16 = const()[name = tensor<string, []>("op_438_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_49_cast = mul(x = var_437_cast, y = var_438_to_fp16)[name = tensor<string, []>("aw_chunk_49_cast")];
            tensor<string, []> var_441_equation_0 = const()[name = tensor<string, []>("op_441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_441_cast = einsum(equation = var_441_equation_0, values = (var_315_cast, var_284_cast))[name = tensor<string, []>("op_441_cast")];
            tensor<fp16, []> var_442_to_fp16 = const()[name = tensor<string, []>("op_442_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_51_cast = mul(x = var_441_cast, y = var_442_to_fp16)[name = tensor<string, []>("aw_chunk_51_cast")];
            tensor<string, []> var_445_equation_0 = const()[name = tensor<string, []>("op_445_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_445_cast = einsum(equation = var_445_equation_0, values = (var_315_cast, var_285_cast))[name = tensor<string, []>("op_445_cast")];
            tensor<fp16, []> var_446_to_fp16 = const()[name = tensor<string, []>("op_446_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_53_cast = mul(x = var_445_cast, y = var_446_to_fp16)[name = tensor<string, []>("aw_chunk_53_cast")];
            tensor<string, []> var_449_equation_0 = const()[name = tensor<string, []>("op_449_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_449_cast = einsum(equation = var_449_equation_0, values = (var_315_cast, var_286_cast))[name = tensor<string, []>("op_449_cast")];
            tensor<fp16, []> var_450_to_fp16 = const()[name = tensor<string, []>("op_450_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_55_cast = mul(x = var_449_cast, y = var_450_to_fp16)[name = tensor<string, []>("aw_chunk_55_cast")];
            tensor<string, []> var_453_equation_0 = const()[name = tensor<string, []>("op_453_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_453_cast = einsum(equation = var_453_equation_0, values = (var_315_cast, var_287_cast))[name = tensor<string, []>("op_453_cast")];
            tensor<fp16, []> var_454_to_fp16 = const()[name = tensor<string, []>("op_454_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_57_cast = mul(x = var_453_cast, y = var_454_to_fp16)[name = tensor<string, []>("aw_chunk_57_cast")];
            tensor<string, []> var_457_equation_0 = const()[name = tensor<string, []>("op_457_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_457_cast = einsum(equation = var_457_equation_0, values = (var_315_cast, var_288_cast))[name = tensor<string, []>("op_457_cast")];
            tensor<fp16, []> var_458_to_fp16 = const()[name = tensor<string, []>("op_458_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_59_cast = mul(x = var_457_cast, y = var_458_to_fp16)[name = tensor<string, []>("aw_chunk_59_cast")];
            tensor<string, []> var_461_equation_0 = const()[name = tensor<string, []>("op_461_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_461_cast = einsum(equation = var_461_equation_0, values = (var_315_cast, var_289_cast))[name = tensor<string, []>("op_461_cast")];
            tensor<fp16, []> var_462_to_fp16 = const()[name = tensor<string, []>("op_462_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_61_cast = mul(x = var_461_cast, y = var_462_to_fp16)[name = tensor<string, []>("aw_chunk_61_cast")];
            tensor<string, []> var_465_equation_0 = const()[name = tensor<string, []>("op_465_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_465_cast = einsum(equation = var_465_equation_0, values = (var_315_cast, var_290_cast))[name = tensor<string, []>("op_465_cast")];
            tensor<fp16, []> var_466_to_fp16 = const()[name = tensor<string, []>("op_466_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_63_cast = mul(x = var_465_cast, y = var_466_to_fp16)[name = tensor<string, []>("aw_chunk_63_cast")];
            tensor<string, []> var_469_equation_0 = const()[name = tensor<string, []>("op_469_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_469_cast = einsum(equation = var_469_equation_0, values = (var_319_cast, var_291_cast))[name = tensor<string, []>("op_469_cast")];
            tensor<fp16, []> var_470_to_fp16 = const()[name = tensor<string, []>("op_470_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_65_cast = mul(x = var_469_cast, y = var_470_to_fp16)[name = tensor<string, []>("aw_chunk_65_cast")];
            tensor<string, []> var_473_equation_0 = const()[name = tensor<string, []>("op_473_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_473_cast = einsum(equation = var_473_equation_0, values = (var_319_cast, var_292_cast))[name = tensor<string, []>("op_473_cast")];
            tensor<fp16, []> var_474_to_fp16 = const()[name = tensor<string, []>("op_474_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_67_cast = mul(x = var_473_cast, y = var_474_to_fp16)[name = tensor<string, []>("aw_chunk_67_cast")];
            tensor<string, []> var_477_equation_0 = const()[name = tensor<string, []>("op_477_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_477_cast = einsum(equation = var_477_equation_0, values = (var_319_cast, var_293_cast))[name = tensor<string, []>("op_477_cast")];
            tensor<fp16, []> var_478_to_fp16 = const()[name = tensor<string, []>("op_478_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_69_cast = mul(x = var_477_cast, y = var_478_to_fp16)[name = tensor<string, []>("aw_chunk_69_cast")];
            tensor<string, []> var_481_equation_0 = const()[name = tensor<string, []>("op_481_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_481_cast = einsum(equation = var_481_equation_0, values = (var_319_cast, var_294_cast))[name = tensor<string, []>("op_481_cast")];
            tensor<fp16, []> var_482_to_fp16 = const()[name = tensor<string, []>("op_482_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_71_cast = mul(x = var_481_cast, y = var_482_to_fp16)[name = tensor<string, []>("aw_chunk_71_cast")];
            tensor<string, []> var_485_equation_0 = const()[name = tensor<string, []>("op_485_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_485_cast = einsum(equation = var_485_equation_0, values = (var_319_cast, var_295_cast))[name = tensor<string, []>("op_485_cast")];
            tensor<fp16, []> var_486_to_fp16 = const()[name = tensor<string, []>("op_486_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_73_cast = mul(x = var_485_cast, y = var_486_to_fp16)[name = tensor<string, []>("aw_chunk_73_cast")];
            tensor<string, []> var_489_equation_0 = const()[name = tensor<string, []>("op_489_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_489_cast = einsum(equation = var_489_equation_0, values = (var_319_cast, var_296_cast))[name = tensor<string, []>("op_489_cast")];
            tensor<fp16, []> var_490_to_fp16 = const()[name = tensor<string, []>("op_490_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_75_cast = mul(x = var_489_cast, y = var_490_to_fp16)[name = tensor<string, []>("aw_chunk_75_cast")];
            tensor<string, []> var_493_equation_0 = const()[name = tensor<string, []>("op_493_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_493_cast = einsum(equation = var_493_equation_0, values = (var_319_cast, var_297_cast))[name = tensor<string, []>("op_493_cast")];
            tensor<fp16, []> var_494_to_fp16 = const()[name = tensor<string, []>("op_494_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_77_cast = mul(x = var_493_cast, y = var_494_to_fp16)[name = tensor<string, []>("aw_chunk_77_cast")];
            tensor<string, []> var_497_equation_0 = const()[name = tensor<string, []>("op_497_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_497_cast = einsum(equation = var_497_equation_0, values = (var_319_cast, var_298_cast))[name = tensor<string, []>("op_497_cast")];
            tensor<fp16, []> var_498_to_fp16 = const()[name = tensor<string, []>("op_498_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_79_cast = mul(x = var_497_cast, y = var_498_to_fp16)[name = tensor<string, []>("aw_chunk_79_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_500_cast = softmax(axis = var_120, x = aw_chunk_1_cast)[name = tensor<string, []>("op_500_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_501_cast = softmax(axis = var_120, x = aw_chunk_3_cast)[name = tensor<string, []>("op_501_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_502_cast = softmax(axis = var_120, x = aw_chunk_5_cast)[name = tensor<string, []>("op_502_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_503_cast = softmax(axis = var_120, x = aw_chunk_7_cast)[name = tensor<string, []>("op_503_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_504_cast = softmax(axis = var_120, x = aw_chunk_9_cast)[name = tensor<string, []>("op_504_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_505_cast = softmax(axis = var_120, x = aw_chunk_11_cast)[name = tensor<string, []>("op_505_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_506_cast = softmax(axis = var_120, x = aw_chunk_13_cast)[name = tensor<string, []>("op_506_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_507_cast = softmax(axis = var_120, x = aw_chunk_15_cast)[name = tensor<string, []>("op_507_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_508_cast = softmax(axis = var_120, x = aw_chunk_17_cast)[name = tensor<string, []>("op_508_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_509_cast = softmax(axis = var_120, x = aw_chunk_19_cast)[name = tensor<string, []>("op_509_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_510_cast = softmax(axis = var_120, x = aw_chunk_21_cast)[name = tensor<string, []>("op_510_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_511_cast = softmax(axis = var_120, x = aw_chunk_23_cast)[name = tensor<string, []>("op_511_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_512_cast = softmax(axis = var_120, x = aw_chunk_25_cast)[name = tensor<string, []>("op_512_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_513_cast = softmax(axis = var_120, x = aw_chunk_27_cast)[name = tensor<string, []>("op_513_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_514_cast = softmax(axis = var_120, x = aw_chunk_29_cast)[name = tensor<string, []>("op_514_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_515_cast = softmax(axis = var_120, x = aw_chunk_31_cast)[name = tensor<string, []>("op_515_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_516_cast = softmax(axis = var_120, x = aw_chunk_33_cast)[name = tensor<string, []>("op_516_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_517_cast = softmax(axis = var_120, x = aw_chunk_35_cast)[name = tensor<string, []>("op_517_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_518_cast = softmax(axis = var_120, x = aw_chunk_37_cast)[name = tensor<string, []>("op_518_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_519_cast = softmax(axis = var_120, x = aw_chunk_39_cast)[name = tensor<string, []>("op_519_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_520_cast = softmax(axis = var_120, x = aw_chunk_41_cast)[name = tensor<string, []>("op_520_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_521_cast = softmax(axis = var_120, x = aw_chunk_43_cast)[name = tensor<string, []>("op_521_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_522_cast = softmax(axis = var_120, x = aw_chunk_45_cast)[name = tensor<string, []>("op_522_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_523_cast = softmax(axis = var_120, x = aw_chunk_47_cast)[name = tensor<string, []>("op_523_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_524_cast = softmax(axis = var_120, x = aw_chunk_49_cast)[name = tensor<string, []>("op_524_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_525_cast = softmax(axis = var_120, x = aw_chunk_51_cast)[name = tensor<string, []>("op_525_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_526_cast = softmax(axis = var_120, x = aw_chunk_53_cast)[name = tensor<string, []>("op_526_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_527_cast = softmax(axis = var_120, x = aw_chunk_55_cast)[name = tensor<string, []>("op_527_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_528_cast = softmax(axis = var_120, x = aw_chunk_57_cast)[name = tensor<string, []>("op_528_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_529_cast = softmax(axis = var_120, x = aw_chunk_59_cast)[name = tensor<string, []>("op_529_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_530_cast = softmax(axis = var_120, x = aw_chunk_61_cast)[name = tensor<string, []>("op_530_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_531_cast = softmax(axis = var_120, x = aw_chunk_63_cast)[name = tensor<string, []>("op_531_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_532_cast = softmax(axis = var_120, x = aw_chunk_65_cast)[name = tensor<string, []>("op_532_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_533_cast = softmax(axis = var_120, x = aw_chunk_67_cast)[name = tensor<string, []>("op_533_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_534_cast = softmax(axis = var_120, x = aw_chunk_69_cast)[name = tensor<string, []>("op_534_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_535_cast = softmax(axis = var_120, x = aw_chunk_71_cast)[name = tensor<string, []>("op_535_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_536_cast = softmax(axis = var_120, x = aw_chunk_73_cast)[name = tensor<string, []>("op_536_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_537_cast = softmax(axis = var_120, x = aw_chunk_75_cast)[name = tensor<string, []>("op_537_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_538_cast = softmax(axis = var_120, x = aw_chunk_77_cast)[name = tensor<string, []>("op_538_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_539_cast = softmax(axis = var_120, x = aw_chunk_79_cast)[name = tensor<string, []>("op_539_cast")];
            tensor<string, []> var_541_equation_0 = const()[name = tensor<string, []>("op_541_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_541_cast = einsum(equation = var_541_equation_0, values = (var_321_cast, var_500_cast))[name = tensor<string, []>("op_541_cast")];
            tensor<string, []> var_543_equation_0 = const()[name = tensor<string, []>("op_543_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_543_cast = einsum(equation = var_543_equation_0, values = (var_321_cast, var_501_cast))[name = tensor<string, []>("op_543_cast")];
            tensor<string, []> var_545_equation_0 = const()[name = tensor<string, []>("op_545_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_545_cast = einsum(equation = var_545_equation_0, values = (var_321_cast, var_502_cast))[name = tensor<string, []>("op_545_cast")];
            tensor<string, []> var_547_equation_0 = const()[name = tensor<string, []>("op_547_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_547_cast = einsum(equation = var_547_equation_0, values = (var_321_cast, var_503_cast))[name = tensor<string, []>("op_547_cast")];
            tensor<string, []> var_549_equation_0 = const()[name = tensor<string, []>("op_549_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_549_cast = einsum(equation = var_549_equation_0, values = (var_321_cast, var_504_cast))[name = tensor<string, []>("op_549_cast")];
            tensor<string, []> var_551_equation_0 = const()[name = tensor<string, []>("op_551_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_551_cast = einsum(equation = var_551_equation_0, values = (var_321_cast, var_505_cast))[name = tensor<string, []>("op_551_cast")];
            tensor<string, []> var_553_equation_0 = const()[name = tensor<string, []>("op_553_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_553_cast = einsum(equation = var_553_equation_0, values = (var_321_cast, var_506_cast))[name = tensor<string, []>("op_553_cast")];
            tensor<string, []> var_555_equation_0 = const()[name = tensor<string, []>("op_555_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_555_cast = einsum(equation = var_555_equation_0, values = (var_321_cast, var_507_cast))[name = tensor<string, []>("op_555_cast")];
            tensor<string, []> var_557_equation_0 = const()[name = tensor<string, []>("op_557_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_557_cast = einsum(equation = var_557_equation_0, values = (var_325_cast, var_508_cast))[name = tensor<string, []>("op_557_cast")];
            tensor<string, []> var_559_equation_0 = const()[name = tensor<string, []>("op_559_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_559_cast = einsum(equation = var_559_equation_0, values = (var_325_cast, var_509_cast))[name = tensor<string, []>("op_559_cast")];
            tensor<string, []> var_561_equation_0 = const()[name = tensor<string, []>("op_561_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_561_cast = einsum(equation = var_561_equation_0, values = (var_325_cast, var_510_cast))[name = tensor<string, []>("op_561_cast")];
            tensor<string, []> var_563_equation_0 = const()[name = tensor<string, []>("op_563_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_563_cast = einsum(equation = var_563_equation_0, values = (var_325_cast, var_511_cast))[name = tensor<string, []>("op_563_cast")];
            tensor<string, []> var_565_equation_0 = const()[name = tensor<string, []>("op_565_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_565_cast = einsum(equation = var_565_equation_0, values = (var_325_cast, var_512_cast))[name = tensor<string, []>("op_565_cast")];
            tensor<string, []> var_567_equation_0 = const()[name = tensor<string, []>("op_567_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_567_cast = einsum(equation = var_567_equation_0, values = (var_325_cast, var_513_cast))[name = tensor<string, []>("op_567_cast")];
            tensor<string, []> var_569_equation_0 = const()[name = tensor<string, []>("op_569_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_569_cast = einsum(equation = var_569_equation_0, values = (var_325_cast, var_514_cast))[name = tensor<string, []>("op_569_cast")];
            tensor<string, []> var_571_equation_0 = const()[name = tensor<string, []>("op_571_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_571_cast = einsum(equation = var_571_equation_0, values = (var_325_cast, var_515_cast))[name = tensor<string, []>("op_571_cast")];
            tensor<string, []> var_573_equation_0 = const()[name = tensor<string, []>("op_573_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_573_cast = einsum(equation = var_573_equation_0, values = (var_329_cast, var_516_cast))[name = tensor<string, []>("op_573_cast")];
            tensor<string, []> var_575_equation_0 = const()[name = tensor<string, []>("op_575_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_575_cast = einsum(equation = var_575_equation_0, values = (var_329_cast, var_517_cast))[name = tensor<string, []>("op_575_cast")];
            tensor<string, []> var_577_equation_0 = const()[name = tensor<string, []>("op_577_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_577_cast = einsum(equation = var_577_equation_0, values = (var_329_cast, var_518_cast))[name = tensor<string, []>("op_577_cast")];
            tensor<string, []> var_579_equation_0 = const()[name = tensor<string, []>("op_579_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_579_cast = einsum(equation = var_579_equation_0, values = (var_329_cast, var_519_cast))[name = tensor<string, []>("op_579_cast")];
            tensor<string, []> var_581_equation_0 = const()[name = tensor<string, []>("op_581_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_581_cast = einsum(equation = var_581_equation_0, values = (var_329_cast, var_520_cast))[name = tensor<string, []>("op_581_cast")];
            tensor<string, []> var_583_equation_0 = const()[name = tensor<string, []>("op_583_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_583_cast = einsum(equation = var_583_equation_0, values = (var_329_cast, var_521_cast))[name = tensor<string, []>("op_583_cast")];
            tensor<string, []> var_585_equation_0 = const()[name = tensor<string, []>("op_585_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_585_cast = einsum(equation = var_585_equation_0, values = (var_329_cast, var_522_cast))[name = tensor<string, []>("op_585_cast")];
            tensor<string, []> var_587_equation_0 = const()[name = tensor<string, []>("op_587_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_587_cast = einsum(equation = var_587_equation_0, values = (var_329_cast, var_523_cast))[name = tensor<string, []>("op_587_cast")];
            tensor<string, []> var_589_equation_0 = const()[name = tensor<string, []>("op_589_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_589_cast = einsum(equation = var_589_equation_0, values = (var_333_cast, var_524_cast))[name = tensor<string, []>("op_589_cast")];
            tensor<string, []> var_591_equation_0 = const()[name = tensor<string, []>("op_591_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_591_cast = einsum(equation = var_591_equation_0, values = (var_333_cast, var_525_cast))[name = tensor<string, []>("op_591_cast")];
            tensor<string, []> var_593_equation_0 = const()[name = tensor<string, []>("op_593_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_593_cast = einsum(equation = var_593_equation_0, values = (var_333_cast, var_526_cast))[name = tensor<string, []>("op_593_cast")];
            tensor<string, []> var_595_equation_0 = const()[name = tensor<string, []>("op_595_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_595_cast = einsum(equation = var_595_equation_0, values = (var_333_cast, var_527_cast))[name = tensor<string, []>("op_595_cast")];
            tensor<string, []> var_597_equation_0 = const()[name = tensor<string, []>("op_597_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_597_cast = einsum(equation = var_597_equation_0, values = (var_333_cast, var_528_cast))[name = tensor<string, []>("op_597_cast")];
            tensor<string, []> var_599_equation_0 = const()[name = tensor<string, []>("op_599_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_599_cast = einsum(equation = var_599_equation_0, values = (var_333_cast, var_529_cast))[name = tensor<string, []>("op_599_cast")];
            tensor<string, []> var_601_equation_0 = const()[name = tensor<string, []>("op_601_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_601_cast = einsum(equation = var_601_equation_0, values = (var_333_cast, var_530_cast))[name = tensor<string, []>("op_601_cast")];
            tensor<string, []> var_603_equation_0 = const()[name = tensor<string, []>("op_603_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_603_cast = einsum(equation = var_603_equation_0, values = (var_333_cast, var_531_cast))[name = tensor<string, []>("op_603_cast")];
            tensor<string, []> var_605_equation_0 = const()[name = tensor<string, []>("op_605_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_605_cast = einsum(equation = var_605_equation_0, values = (var_337_cast, var_532_cast))[name = tensor<string, []>("op_605_cast")];
            tensor<string, []> var_607_equation_0 = const()[name = tensor<string, []>("op_607_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_607_cast = einsum(equation = var_607_equation_0, values = (var_337_cast, var_533_cast))[name = tensor<string, []>("op_607_cast")];
            tensor<string, []> var_609_equation_0 = const()[name = tensor<string, []>("op_609_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_609_cast = einsum(equation = var_609_equation_0, values = (var_337_cast, var_534_cast))[name = tensor<string, []>("op_609_cast")];
            tensor<string, []> var_611_equation_0 = const()[name = tensor<string, []>("op_611_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_611_cast = einsum(equation = var_611_equation_0, values = (var_337_cast, var_535_cast))[name = tensor<string, []>("op_611_cast")];
            tensor<string, []> var_613_equation_0 = const()[name = tensor<string, []>("op_613_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_613_cast = einsum(equation = var_613_equation_0, values = (var_337_cast, var_536_cast))[name = tensor<string, []>("op_613_cast")];
            tensor<string, []> var_615_equation_0 = const()[name = tensor<string, []>("op_615_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_615_cast = einsum(equation = var_615_equation_0, values = (var_337_cast, var_537_cast))[name = tensor<string, []>("op_615_cast")];
            tensor<string, []> var_617_equation_0 = const()[name = tensor<string, []>("op_617_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_617_cast = einsum(equation = var_617_equation_0, values = (var_337_cast, var_538_cast))[name = tensor<string, []>("op_617_cast")];
            tensor<string, []> var_619_equation_0 = const()[name = tensor<string, []>("op_619_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_619_cast = einsum(equation = var_619_equation_0, values = (var_337_cast, var_539_cast))[name = tensor<string, []>("op_619_cast")];
            tensor<bool, []> var_621_interleave_0 = const()[name = tensor<string, []>("op_621_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_621_cast = concat(axis = var_95, interleave = var_621_interleave_0, values = (var_541_cast, var_543_cast, var_545_cast, var_547_cast, var_549_cast, var_551_cast, var_553_cast, var_555_cast))[name = tensor<string, []>("op_621_cast")];
            tensor<bool, []> var_623_interleave_0 = const()[name = tensor<string, []>("op_623_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_623_cast = concat(axis = var_95, interleave = var_623_interleave_0, values = (var_557_cast, var_559_cast, var_561_cast, var_563_cast, var_565_cast, var_567_cast, var_569_cast, var_571_cast))[name = tensor<string, []>("op_623_cast")];
            tensor<bool, []> var_625_interleave_0 = const()[name = tensor<string, []>("op_625_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_625_cast = concat(axis = var_95, interleave = var_625_interleave_0, values = (var_573_cast, var_575_cast, var_577_cast, var_579_cast, var_581_cast, var_583_cast, var_585_cast, var_587_cast))[name = tensor<string, []>("op_625_cast")];
            tensor<bool, []> var_627_interleave_0 = const()[name = tensor<string, []>("op_627_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_627_cast = concat(axis = var_95, interleave = var_627_interleave_0, values = (var_589_cast, var_591_cast, var_593_cast, var_595_cast, var_597_cast, var_599_cast, var_601_cast, var_603_cast))[name = tensor<string, []>("op_627_cast")];
            tensor<bool, []> var_629_interleave_0 = const()[name = tensor<string, []>("op_629_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_629_cast = concat(axis = var_95, interleave = var_629_interleave_0, values = (var_605_cast, var_607_cast, var_609_cast, var_611_cast, var_613_cast, var_615_cast, var_617_cast, var_619_cast))[name = tensor<string, []>("op_629_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_120, interleave = input_25_interleave_0, values = (var_621_cast, var_623_cast, var_625_cast, var_627_cast, var_629_cast))[name = tensor<string, []>("input_25_cast")];
            tensor<int32, [2]> var_635 = const()[name = tensor<string, []>("op_635"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_637 = const()[name = tensor<string, []>("op_637"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_639_pad_type_0 = const()[name = tensor<string, []>("op_639_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_639_pad_0 = const()[name = tensor<string, []>("op_639_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1197376))), lut = tensor<fp16, [4]>([-0x1.3bcp-4, -0x1.714p-6, 0x1.7d4p-6, 0x1.3f8p-4]), 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, []>(1223040)))];
            tensor<fp16, [2, 320, 1, 4096]> var_639_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_637, groups = var_120, pad = var_639_pad_0, pad_type = var_639_pad_type_0, strides = var_635, 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_639_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_3_cast = add(x = var_639_cast, y = inputs_1_cast)[name = tensor<string, []>("inputs_3_cast")];
            tensor<int32, [1]> var_643 = const()[name = tensor<string, []>("op_643"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_3_cast = reduce_mean(axes = var_643, keep_dims = var_115, 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_647 = const()[name = tensor<string, []>("op_647"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_648_cast = reduce_mean(axes = var_647, keep_dims = var_115, x = zero_mean_sq_3_cast)[name = tensor<string, []>("op_648_cast")];
            tensor<fp16, []> var_649_to_fp16 = const()[name = tensor<string, []>("op_649_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_650_cast = add(x = var_648_cast, y = var_649_to_fp16)[name = tensor<string, []>("op_650_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_650_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_654_to_fp16 = const()[name = tensor<string, []>("op_654_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1223744)))];
            tensor<fp16, [2, 320, 1, 4096]> var_655_cast = add(x = out_3_cast, y = var_654_to_fp16)[name = tensor<string, []>("op_655_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_657_to_fp16 = const()[name = tensor<string, []>("op_657_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1224448)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_11_cast = mul(x = var_655_cast, y = var_657_to_fp16)[name = tensor<string, []>("hidden_states_11_cast")];
            tensor<int32, [2]> var_664 = const()[name = tensor<string, []>("op_664"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_666 = const()[name = tensor<string, []>("op_666"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1225152))), lut = tensor<fp16, [4]>([-0x1.cacp-4, -0x1.1p-5, 0x1.15p-5, 0x1.cf4p-4]), 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_666, groups = var_120, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = var_664, 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_670 = const()[name = tensor<string, []>("op_670"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_672 = const()[name = tensor<string, []>("op_672"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_5_pad_type_0 = const()[name = tensor<string, []>("k_5_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_5_pad_0 = const()[name = tensor<string, []>("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1250816))), lut = tensor<fp16, [4]>([-0x1.82p-4, -0x1.c3cp-6, 0x1.c1p-6, 0x1.814p-4]), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_5_cast = conv(dilations = var_672, groups = var_120, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = var_670, 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_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_678 = const()[name = tensor<string, []>("op_678"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_3_pad_type_0 = const()[name = tensor<string, []>("v_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_3_pad_0 = const()[name = tensor<string, []>("v_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1332800))), lut = tensor<fp16, [4]>([-0x1.c84p-6, -0x1p-7, 0x1.0bcp-7, 0x1.ce4p-6]), name = tensor<string, []>("down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_3_cast = conv(dilations = var_678, groups = var_120, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = var_676, 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_682_begin_0 = const()[name = tensor<string, []>("op_682_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_682_end_0 = const()[name = tensor<string, []>("op_682_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_682_end_mask_0 = const()[name = tensor<string, []>("op_682_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_682_cast = slice_by_index(begin = var_682_begin_0, end = var_682_end_0, end_mask = var_682_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_682_cast")];
            tensor<int32, [4]> var_686_begin_0 = const()[name = tensor<string, []>("op_686_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_686_end_0 = const()[name = tensor<string, []>("op_686_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_686_end_mask_0 = const()[name = tensor<string, []>("op_686_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_686_cast = slice_by_index(begin = var_686_begin_0, end = var_686_end_0, end_mask = var_686_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_686_cast")];
            tensor<int32, [4]> var_690_begin_0 = const()[name = tensor<string, []>("op_690_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_690_end_0 = const()[name = tensor<string, []>("op_690_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_690_end_mask_0 = const()[name = tensor<string, []>("op_690_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_690_cast = slice_by_index(begin = var_690_begin_0, end = var_690_end_0, end_mask = var_690_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_690_cast")];
            tensor<int32, [4]> var_694_begin_0 = const()[name = tensor<string, []>("op_694_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_694_end_0 = const()[name = tensor<string, []>("op_694_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_694_end_mask_0 = const()[name = tensor<string, []>("op_694_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_694_cast = slice_by_index(begin = var_694_begin_0, end = var_694_end_0, end_mask = var_694_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_694_cast")];
            tensor<int32, [4]> var_698_begin_0 = const()[name = tensor<string, []>("op_698_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_698_end_0 = const()[name = tensor<string, []>("op_698_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_698_end_mask_0 = const()[name = tensor<string, []>("op_698_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_698_cast = slice_by_index(begin = var_698_begin_0, end = var_698_end_0, end_mask = var_698_end_mask_0, x = q_3_cast)[name = tensor<string, []>("op_698_cast")];
            tensor<int32, [4]> var_701_begin_0 = const()[name = tensor<string, []>("op_701_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_701_end_0 = const()[name = tensor<string, []>("op_701_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_701_end_mask_0 = const()[name = tensor<string, []>("op_701_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_701_cast = slice_by_index(begin = var_701_begin_0, end = var_701_end_0, end_mask = var_701_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_701_cast")];
            tensor<int32, [4]> var_702_begin_0 = const()[name = tensor<string, []>("op_702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_702_end_0 = const()[name = tensor<string, []>("op_702_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_702_end_mask_0 = const()[name = tensor<string, []>("op_702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_702_cast = slice_by_index(begin = var_702_begin_0, end = var_702_end_0, end_mask = var_702_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_702_cast")];
            tensor<int32, [4]> var_703_begin_0 = const()[name = tensor<string, []>("op_703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_703_end_0 = const()[name = tensor<string, []>("op_703_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_703_end_mask_0 = const()[name = tensor<string, []>("op_703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_703_cast = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_703_cast")];
            tensor<int32, [4]> var_704_begin_0 = const()[name = tensor<string, []>("op_704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_704_end_0 = const()[name = tensor<string, []>("op_704_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_704_end_mask_0 = const()[name = tensor<string, []>("op_704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_704_cast = slice_by_index(begin = var_704_begin_0, end = var_704_end_0, end_mask = var_704_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_704_cast")];
            tensor<int32, [4]> var_705_begin_0 = const()[name = tensor<string, []>("op_705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_705_end_0 = const()[name = tensor<string, []>("op_705_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_705_end_mask_0 = const()[name = tensor<string, []>("op_705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_705_cast = slice_by_index(begin = var_705_begin_0, end = var_705_end_0, end_mask = var_705_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_705_cast")];
            tensor<int32, [4]> var_706_begin_0 = const()[name = tensor<string, []>("op_706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_706_end_0 = const()[name = tensor<string, []>("op_706_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_706_end_mask_0 = const()[name = tensor<string, []>("op_706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_706_cast = slice_by_index(begin = var_706_begin_0, end = var_706_end_0, end_mask = var_706_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_706_cast")];
            tensor<int32, [4]> var_707_begin_0 = const()[name = tensor<string, []>("op_707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_707_end_0 = const()[name = tensor<string, []>("op_707_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_707_end_mask_0 = const()[name = tensor<string, []>("op_707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_707_cast = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_707_cast")];
            tensor<int32, [4]> var_708_begin_0 = const()[name = tensor<string, []>("op_708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_708_end_0 = const()[name = tensor<string, []>("op_708_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_708_end_mask_0 = const()[name = tensor<string, []>("op_708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_708_cast = slice_by_index(begin = var_708_begin_0, end = var_708_end_0, end_mask = var_708_end_mask_0, x = var_682_cast)[name = tensor<string, []>("op_708_cast")];
            tensor<int32, [4]> var_709_begin_0 = const()[name = tensor<string, []>("op_709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_709_end_0 = const()[name = tensor<string, []>("op_709_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_709_end_mask_0 = const()[name = tensor<string, []>("op_709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_709_cast = slice_by_index(begin = var_709_begin_0, end = var_709_end_0, end_mask = var_709_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_709_cast")];
            tensor<int32, [4]> var_710_begin_0 = const()[name = tensor<string, []>("op_710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_710_end_0 = const()[name = tensor<string, []>("op_710_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_710_end_mask_0 = const()[name = tensor<string, []>("op_710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_710_cast = slice_by_index(begin = var_710_begin_0, end = var_710_end_0, end_mask = var_710_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_710_cast")];
            tensor<int32, [4]> var_711_begin_0 = const()[name = tensor<string, []>("op_711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_711_end_0 = const()[name = tensor<string, []>("op_711_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_711_end_mask_0 = const()[name = tensor<string, []>("op_711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_711_cast = slice_by_index(begin = var_711_begin_0, end = var_711_end_0, end_mask = var_711_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_711_cast")];
            tensor<int32, [4]> var_712_begin_0 = const()[name = tensor<string, []>("op_712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_712_end_0 = const()[name = tensor<string, []>("op_712_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_712_end_mask_0 = const()[name = tensor<string, []>("op_712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_712_cast = slice_by_index(begin = var_712_begin_0, end = var_712_end_0, end_mask = var_712_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_712_cast")];
            tensor<int32, [4]> var_713_begin_0 = const()[name = tensor<string, []>("op_713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_713_end_0 = const()[name = tensor<string, []>("op_713_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_713_end_mask_0 = const()[name = tensor<string, []>("op_713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_713_cast = slice_by_index(begin = var_713_begin_0, end = var_713_end_0, end_mask = var_713_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_713_cast")];
            tensor<int32, [4]> var_714_begin_0 = const()[name = tensor<string, []>("op_714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_714_end_0 = const()[name = tensor<string, []>("op_714_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_714_end_mask_0 = const()[name = tensor<string, []>("op_714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_714_cast = slice_by_index(begin = var_714_begin_0, end = var_714_end_0, end_mask = var_714_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_714_cast")];
            tensor<int32, [4]> var_715_begin_0 = const()[name = tensor<string, []>("op_715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_715_end_0 = const()[name = tensor<string, []>("op_715_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_715_end_mask_0 = const()[name = tensor<string, []>("op_715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_715_cast = slice_by_index(begin = var_715_begin_0, end = var_715_end_0, end_mask = var_715_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_715_cast")];
            tensor<int32, [4]> var_716_begin_0 = const()[name = tensor<string, []>("op_716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_716_end_0 = const()[name = tensor<string, []>("op_716_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_716_end_mask_0 = const()[name = tensor<string, []>("op_716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_716_cast = slice_by_index(begin = var_716_begin_0, end = var_716_end_0, end_mask = var_716_end_mask_0, x = var_686_cast)[name = tensor<string, []>("op_716_cast")];
            tensor<int32, [4]> var_717_begin_0 = const()[name = tensor<string, []>("op_717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_717_end_0 = const()[name = tensor<string, []>("op_717_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_717_end_mask_0 = const()[name = tensor<string, []>("op_717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_717_cast = slice_by_index(begin = var_717_begin_0, end = var_717_end_0, end_mask = var_717_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_717_cast")];
            tensor<int32, [4]> var_718_begin_0 = const()[name = tensor<string, []>("op_718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_718_end_0 = const()[name = tensor<string, []>("op_718_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_718_end_mask_0 = const()[name = tensor<string, []>("op_718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_718_cast = slice_by_index(begin = var_718_begin_0, end = var_718_end_0, end_mask = var_718_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_718_cast")];
            tensor<int32, [4]> var_719_begin_0 = const()[name = tensor<string, []>("op_719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_719_end_0 = const()[name = tensor<string, []>("op_719_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_719_end_mask_0 = const()[name = tensor<string, []>("op_719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_719_cast = slice_by_index(begin = var_719_begin_0, end = var_719_end_0, end_mask = var_719_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_719_cast")];
            tensor<int32, [4]> var_720_begin_0 = const()[name = tensor<string, []>("op_720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_720_end_0 = const()[name = tensor<string, []>("op_720_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_720_end_mask_0 = const()[name = tensor<string, []>("op_720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_720_cast = slice_by_index(begin = var_720_begin_0, end = var_720_end_0, end_mask = var_720_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_720_cast")];
            tensor<int32, [4]> var_721_begin_0 = const()[name = tensor<string, []>("op_721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_721_end_0 = const()[name = tensor<string, []>("op_721_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_721_end_mask_0 = const()[name = tensor<string, []>("op_721_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_721_cast = slice_by_index(begin = var_721_begin_0, end = var_721_end_0, end_mask = var_721_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_721_cast")];
            tensor<int32, [4]> var_722_begin_0 = const()[name = tensor<string, []>("op_722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_722_end_0 = const()[name = tensor<string, []>("op_722_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_722_end_mask_0 = const()[name = tensor<string, []>("op_722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_722_cast = slice_by_index(begin = var_722_begin_0, end = var_722_end_0, end_mask = var_722_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_722_cast")];
            tensor<int32, [4]> var_723_begin_0 = const()[name = tensor<string, []>("op_723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_723_end_0 = const()[name = tensor<string, []>("op_723_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_723_end_mask_0 = const()[name = tensor<string, []>("op_723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_723_cast = slice_by_index(begin = var_723_begin_0, end = var_723_end_0, end_mask = var_723_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_723_cast")];
            tensor<int32, [4]> var_724_begin_0 = const()[name = tensor<string, []>("op_724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_724_end_0 = const()[name = tensor<string, []>("op_724_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_724_end_mask_0 = const()[name = tensor<string, []>("op_724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_724_cast = slice_by_index(begin = var_724_begin_0, end = var_724_end_0, end_mask = var_724_end_mask_0, x = var_690_cast)[name = tensor<string, []>("op_724_cast")];
            tensor<int32, [4]> var_725_begin_0 = const()[name = tensor<string, []>("op_725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_725_end_0 = const()[name = tensor<string, []>("op_725_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_725_end_mask_0 = const()[name = tensor<string, []>("op_725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_725_cast = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_725_cast")];
            tensor<int32, [4]> var_726_begin_0 = const()[name = tensor<string, []>("op_726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_726_end_0 = const()[name = tensor<string, []>("op_726_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_726_end_mask_0 = const()[name = tensor<string, []>("op_726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_726_cast = slice_by_index(begin = var_726_begin_0, end = var_726_end_0, end_mask = var_726_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_726_cast")];
            tensor<int32, [4]> var_727_begin_0 = const()[name = tensor<string, []>("op_727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_727_end_0 = const()[name = tensor<string, []>("op_727_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_727_end_mask_0 = const()[name = tensor<string, []>("op_727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_727_cast = slice_by_index(begin = var_727_begin_0, end = var_727_end_0, end_mask = var_727_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_727_cast")];
            tensor<int32, [4]> var_728_begin_0 = const()[name = tensor<string, []>("op_728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_728_end_0 = const()[name = tensor<string, []>("op_728_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_728_end_mask_0 = const()[name = tensor<string, []>("op_728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_728_cast = slice_by_index(begin = var_728_begin_0, end = var_728_end_0, end_mask = var_728_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_728_cast")];
            tensor<int32, [4]> var_729_begin_0 = const()[name = tensor<string, []>("op_729_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_729_end_0 = const()[name = tensor<string, []>("op_729_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_729_end_mask_0 = const()[name = tensor<string, []>("op_729_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_729_cast = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_729_cast")];
            tensor<int32, [4]> var_730_begin_0 = const()[name = tensor<string, []>("op_730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_730_end_0 = const()[name = tensor<string, []>("op_730_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_730_end_mask_0 = const()[name = tensor<string, []>("op_730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_730_cast = slice_by_index(begin = var_730_begin_0, end = var_730_end_0, end_mask = var_730_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_730_cast")];
            tensor<int32, [4]> var_731_begin_0 = const()[name = tensor<string, []>("op_731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_731_end_0 = const()[name = tensor<string, []>("op_731_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_731_end_mask_0 = const()[name = tensor<string, []>("op_731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_731_cast = slice_by_index(begin = var_731_begin_0, end = var_731_end_0, end_mask = var_731_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_731_cast")];
            tensor<int32, [4]> var_732_begin_0 = const()[name = tensor<string, []>("op_732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_732_end_0 = const()[name = tensor<string, []>("op_732_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_732_end_mask_0 = const()[name = tensor<string, []>("op_732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_732_cast = slice_by_index(begin = var_732_begin_0, end = var_732_end_0, end_mask = var_732_end_mask_0, x = var_694_cast)[name = tensor<string, []>("op_732_cast")];
            tensor<int32, [4]> var_733_begin_0 = const()[name = tensor<string, []>("op_733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_733_end_0 = const()[name = tensor<string, []>("op_733_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_733_end_mask_0 = const()[name = tensor<string, []>("op_733_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_733_cast = slice_by_index(begin = var_733_begin_0, end = var_733_end_0, end_mask = var_733_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_733_cast")];
            tensor<int32, [4]> var_734_begin_0 = const()[name = tensor<string, []>("op_734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_734_end_0 = const()[name = tensor<string, []>("op_734_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_734_end_mask_0 = const()[name = tensor<string, []>("op_734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_734_cast = slice_by_index(begin = var_734_begin_0, end = var_734_end_0, end_mask = var_734_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_734_cast")];
            tensor<int32, [4]> var_735_begin_0 = const()[name = tensor<string, []>("op_735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_735_end_0 = const()[name = tensor<string, []>("op_735_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_735_end_mask_0 = const()[name = tensor<string, []>("op_735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_735_cast = slice_by_index(begin = var_735_begin_0, end = var_735_end_0, end_mask = var_735_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_735_cast")];
            tensor<int32, [4]> var_736_begin_0 = const()[name = tensor<string, []>("op_736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_736_end_0 = const()[name = tensor<string, []>("op_736_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_736_end_mask_0 = const()[name = tensor<string, []>("op_736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_736_cast = slice_by_index(begin = var_736_begin_0, end = var_736_end_0, end_mask = var_736_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_736_cast")];
            tensor<int32, [4]> var_737_begin_0 = const()[name = tensor<string, []>("op_737_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_737_end_0 = const()[name = tensor<string, []>("op_737_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_737_end_mask_0 = const()[name = tensor<string, []>("op_737_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_737_cast = slice_by_index(begin = var_737_begin_0, end = var_737_end_0, end_mask = var_737_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_737_cast")];
            tensor<int32, [4]> var_738_begin_0 = const()[name = tensor<string, []>("op_738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_738_end_0 = const()[name = tensor<string, []>("op_738_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_738_end_mask_0 = const()[name = tensor<string, []>("op_738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_738_cast = slice_by_index(begin = var_738_begin_0, end = var_738_end_0, end_mask = var_738_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_738_cast")];
            tensor<int32, [4]> var_739_begin_0 = const()[name = tensor<string, []>("op_739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_739_end_0 = const()[name = tensor<string, []>("op_739_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_739_end_mask_0 = const()[name = tensor<string, []>("op_739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_739_cast = slice_by_index(begin = var_739_begin_0, end = var_739_end_0, end_mask = var_739_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_739_cast")];
            tensor<int32, [4]> var_740_begin_0 = const()[name = tensor<string, []>("op_740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_740_end_0 = const()[name = tensor<string, []>("op_740_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_740_end_mask_0 = const()[name = tensor<string, []>("op_740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_740_cast = slice_by_index(begin = var_740_begin_0, end = var_740_end_0, end_mask = var_740_end_mask_0, x = var_698_cast)[name = tensor<string, []>("op_740_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_745_begin_0 = const()[name = tensor<string, []>("op_745_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_745_end_0 = const()[name = tensor<string, []>("op_745_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_745_end_mask_0 = const()[name = tensor<string, []>("op_745_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, 64]> var_745_cast = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_745_cast")];
            tensor<int32, [4]> var_749_begin_0 = const()[name = tensor<string, []>("op_749_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_749_end_0 = const()[name = tensor<string, []>("op_749_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_749_end_mask_0 = const()[name = tensor<string, []>("op_749_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_749_cast = slice_by_index(begin = var_749_begin_0, end = var_749_end_0, end_mask = var_749_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_749_cast")];
            tensor<int32, [4]> var_753_begin_0 = const()[name = tensor<string, []>("op_753_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_753_end_0 = const()[name = tensor<string, []>("op_753_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_753_end_mask_0 = const()[name = tensor<string, []>("op_753_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_753_cast = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_753_cast")];
            tensor<int32, [4]> var_757_begin_0 = const()[name = tensor<string, []>("op_757_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_757_end_0 = const()[name = tensor<string, []>("op_757_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_757_end_mask_0 = const()[name = tensor<string, []>("op_757_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_757_cast = slice_by_index(begin = var_757_begin_0, end = var_757_end_0, end_mask = var_757_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_757_cast")];
            tensor<int32, [4]> var_761_begin_0 = const()[name = tensor<string, []>("op_761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_761_end_0 = const()[name = tensor<string, []>("op_761_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_761_end_mask_0 = const()[name = tensor<string, []>("op_761_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_761_cast = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = transpose_30)[name = tensor<string, []>("op_761_cast")];
            tensor<int32, [4]> var_763_begin_0 = const()[name = tensor<string, []>("op_763_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_763_end_0 = const()[name = tensor<string, []>("op_763_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_763_end_mask_0 = const()[name = tensor<string, []>("op_763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_763_cast = slice_by_index(begin = var_763_begin_0, end = var_763_end_0, end_mask = var_763_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_763_cast")];
            tensor<int32, [4]> var_767_begin_0 = const()[name = tensor<string, []>("op_767_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_767_end_0 = const()[name = tensor<string, []>("op_767_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_767_end_mask_0 = const()[name = tensor<string, []>("op_767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_767_cast = slice_by_index(begin = var_767_begin_0, end = var_767_end_0, end_mask = var_767_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_767_cast")];
            tensor<int32, [4]> var_771_begin_0 = const()[name = tensor<string, []>("op_771_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_771_end_0 = const()[name = tensor<string, []>("op_771_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_771_end_mask_0 = const()[name = tensor<string, []>("op_771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_771_cast = slice_by_index(begin = var_771_begin_0, end = var_771_end_0, end_mask = var_771_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_771_cast")];
            tensor<int32, [4]> var_775_begin_0 = const()[name = tensor<string, []>("op_775_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_775_end_0 = const()[name = tensor<string, []>("op_775_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_775_end_mask_0 = const()[name = tensor<string, []>("op_775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_775_cast = slice_by_index(begin = var_775_begin_0, end = var_775_end_0, end_mask = var_775_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_775_cast")];
            tensor<int32, [4]> var_779_begin_0 = const()[name = tensor<string, []>("op_779_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_779_end_0 = const()[name = tensor<string, []>("op_779_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_779_end_mask_0 = const()[name = tensor<string, []>("op_779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_779_cast = slice_by_index(begin = var_779_begin_0, end = var_779_end_0, end_mask = var_779_end_mask_0, x = v_3_cast)[name = tensor<string, []>("op_779_cast")];
            tensor<string, []> var_783_equation_0 = const()[name = tensor<string, []>("op_783_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_783_cast = einsum(equation = var_783_equation_0, values = (var_745_cast, var_701_cast))[name = tensor<string, []>("op_783_cast")];
            tensor<fp16, []> var_784_to_fp16 = const()[name = tensor<string, []>("op_784_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_81_cast = mul(x = var_783_cast, y = var_784_to_fp16)[name = tensor<string, []>("aw_chunk_81_cast")];
            tensor<string, []> var_787_equation_0 = const()[name = tensor<string, []>("op_787_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_787_cast = einsum(equation = var_787_equation_0, values = (var_745_cast, var_702_cast))[name = tensor<string, []>("op_787_cast")];
            tensor<fp16, []> var_788_to_fp16 = const()[name = tensor<string, []>("op_788_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_83_cast = mul(x = var_787_cast, y = var_788_to_fp16)[name = tensor<string, []>("aw_chunk_83_cast")];
            tensor<string, []> var_791_equation_0 = const()[name = tensor<string, []>("op_791_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_791_cast = einsum(equation = var_791_equation_0, values = (var_745_cast, var_703_cast))[name = tensor<string, []>("op_791_cast")];
            tensor<fp16, []> var_792_to_fp16 = const()[name = tensor<string, []>("op_792_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_85_cast = mul(x = var_791_cast, y = var_792_to_fp16)[name = tensor<string, []>("aw_chunk_85_cast")];
            tensor<string, []> var_795_equation_0 = const()[name = tensor<string, []>("op_795_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_795_cast = einsum(equation = var_795_equation_0, values = (var_745_cast, var_704_cast))[name = tensor<string, []>("op_795_cast")];
            tensor<fp16, []> var_796_to_fp16 = const()[name = tensor<string, []>("op_796_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_87_cast = mul(x = var_795_cast, y = var_796_to_fp16)[name = tensor<string, []>("aw_chunk_87_cast")];
            tensor<string, []> var_799_equation_0 = const()[name = tensor<string, []>("op_799_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_799_cast = einsum(equation = var_799_equation_0, values = (var_745_cast, var_705_cast))[name = tensor<string, []>("op_799_cast")];
            tensor<fp16, []> var_800_to_fp16 = const()[name = tensor<string, []>("op_800_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_89_cast = mul(x = var_799_cast, y = var_800_to_fp16)[name = tensor<string, []>("aw_chunk_89_cast")];
            tensor<string, []> var_803_equation_0 = const()[name = tensor<string, []>("op_803_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_803_cast = einsum(equation = var_803_equation_0, values = (var_745_cast, var_706_cast))[name = tensor<string, []>("op_803_cast")];
            tensor<fp16, []> var_804_to_fp16 = const()[name = tensor<string, []>("op_804_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_91_cast = mul(x = var_803_cast, y = var_804_to_fp16)[name = tensor<string, []>("aw_chunk_91_cast")];
            tensor<string, []> var_807_equation_0 = const()[name = tensor<string, []>("op_807_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_807_cast = einsum(equation = var_807_equation_0, values = (var_745_cast, var_707_cast))[name = tensor<string, []>("op_807_cast")];
            tensor<fp16, []> var_808_to_fp16 = const()[name = tensor<string, []>("op_808_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_93_cast = mul(x = var_807_cast, y = var_808_to_fp16)[name = tensor<string, []>("aw_chunk_93_cast")];
            tensor<string, []> var_811_equation_0 = const()[name = tensor<string, []>("op_811_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_811_cast = einsum(equation = var_811_equation_0, values = (var_745_cast, var_708_cast))[name = tensor<string, []>("op_811_cast")];
            tensor<fp16, []> var_812_to_fp16 = const()[name = tensor<string, []>("op_812_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_95_cast = mul(x = var_811_cast, y = var_812_to_fp16)[name = tensor<string, []>("aw_chunk_95_cast")];
            tensor<string, []> var_815_equation_0 = const()[name = tensor<string, []>("op_815_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_815_cast = einsum(equation = var_815_equation_0, values = (var_749_cast, var_709_cast))[name = tensor<string, []>("op_815_cast")];
            tensor<fp16, []> var_816_to_fp16 = const()[name = tensor<string, []>("op_816_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_97_cast = mul(x = var_815_cast, y = var_816_to_fp16)[name = tensor<string, []>("aw_chunk_97_cast")];
            tensor<string, []> var_819_equation_0 = const()[name = tensor<string, []>("op_819_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_819_cast = einsum(equation = var_819_equation_0, values = (var_749_cast, var_710_cast))[name = tensor<string, []>("op_819_cast")];
            tensor<fp16, []> var_820_to_fp16 = const()[name = tensor<string, []>("op_820_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_99_cast = mul(x = var_819_cast, y = var_820_to_fp16)[name = tensor<string, []>("aw_chunk_99_cast")];
            tensor<string, []> var_823_equation_0 = const()[name = tensor<string, []>("op_823_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_823_cast = einsum(equation = var_823_equation_0, values = (var_749_cast, var_711_cast))[name = tensor<string, []>("op_823_cast")];
            tensor<fp16, []> var_824_to_fp16 = const()[name = tensor<string, []>("op_824_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_101_cast = mul(x = var_823_cast, y = var_824_to_fp16)[name = tensor<string, []>("aw_chunk_101_cast")];
            tensor<string, []> var_827_equation_0 = const()[name = tensor<string, []>("op_827_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_827_cast = einsum(equation = var_827_equation_0, values = (var_749_cast, var_712_cast))[name = tensor<string, []>("op_827_cast")];
            tensor<fp16, []> var_828_to_fp16 = const()[name = tensor<string, []>("op_828_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_103_cast = mul(x = var_827_cast, y = var_828_to_fp16)[name = tensor<string, []>("aw_chunk_103_cast")];
            tensor<string, []> var_831_equation_0 = const()[name = tensor<string, []>("op_831_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_831_cast = einsum(equation = var_831_equation_0, values = (var_749_cast, var_713_cast))[name = tensor<string, []>("op_831_cast")];
            tensor<fp16, []> var_832_to_fp16 = const()[name = tensor<string, []>("op_832_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_105_cast = mul(x = var_831_cast, y = var_832_to_fp16)[name = tensor<string, []>("aw_chunk_105_cast")];
            tensor<string, []> var_835_equation_0 = const()[name = tensor<string, []>("op_835_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_835_cast = einsum(equation = var_835_equation_0, values = (var_749_cast, var_714_cast))[name = tensor<string, []>("op_835_cast")];
            tensor<fp16, []> var_836_to_fp16 = const()[name = tensor<string, []>("op_836_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_107_cast = mul(x = var_835_cast, y = var_836_to_fp16)[name = tensor<string, []>("aw_chunk_107_cast")];
            tensor<string, []> var_839_equation_0 = const()[name = tensor<string, []>("op_839_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_839_cast = einsum(equation = var_839_equation_0, values = (var_749_cast, var_715_cast))[name = tensor<string, []>("op_839_cast")];
            tensor<fp16, []> var_840_to_fp16 = const()[name = tensor<string, []>("op_840_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_109_cast = mul(x = var_839_cast, y = var_840_to_fp16)[name = tensor<string, []>("aw_chunk_109_cast")];
            tensor<string, []> var_843_equation_0 = const()[name = tensor<string, []>("op_843_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_843_cast = einsum(equation = var_843_equation_0, values = (var_749_cast, var_716_cast))[name = tensor<string, []>("op_843_cast")];
            tensor<fp16, []> var_844_to_fp16 = const()[name = tensor<string, []>("op_844_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_111_cast = mul(x = var_843_cast, y = var_844_to_fp16)[name = tensor<string, []>("aw_chunk_111_cast")];
            tensor<string, []> var_847_equation_0 = const()[name = tensor<string, []>("op_847_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_847_cast = einsum(equation = var_847_equation_0, values = (var_753_cast, var_717_cast))[name = tensor<string, []>("op_847_cast")];
            tensor<fp16, []> var_848_to_fp16 = const()[name = tensor<string, []>("op_848_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_113_cast = mul(x = var_847_cast, y = var_848_to_fp16)[name = tensor<string, []>("aw_chunk_113_cast")];
            tensor<string, []> var_851_equation_0 = const()[name = tensor<string, []>("op_851_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_851_cast = einsum(equation = var_851_equation_0, values = (var_753_cast, var_718_cast))[name = tensor<string, []>("op_851_cast")];
            tensor<fp16, []> var_852_to_fp16 = const()[name = tensor<string, []>("op_852_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_115_cast = mul(x = var_851_cast, y = var_852_to_fp16)[name = tensor<string, []>("aw_chunk_115_cast")];
            tensor<string, []> var_855_equation_0 = const()[name = tensor<string, []>("op_855_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_855_cast = einsum(equation = var_855_equation_0, values = (var_753_cast, var_719_cast))[name = tensor<string, []>("op_855_cast")];
            tensor<fp16, []> var_856_to_fp16 = const()[name = tensor<string, []>("op_856_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_117_cast = mul(x = var_855_cast, y = var_856_to_fp16)[name = tensor<string, []>("aw_chunk_117_cast")];
            tensor<string, []> var_859_equation_0 = const()[name = tensor<string, []>("op_859_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_859_cast = einsum(equation = var_859_equation_0, values = (var_753_cast, var_720_cast))[name = tensor<string, []>("op_859_cast")];
            tensor<fp16, []> var_860_to_fp16 = const()[name = tensor<string, []>("op_860_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_119_cast = mul(x = var_859_cast, y = var_860_to_fp16)[name = tensor<string, []>("aw_chunk_119_cast")];
            tensor<string, []> var_863_equation_0 = const()[name = tensor<string, []>("op_863_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_863_cast = einsum(equation = var_863_equation_0, values = (var_753_cast, var_721_cast))[name = tensor<string, []>("op_863_cast")];
            tensor<fp16, []> var_864_to_fp16 = const()[name = tensor<string, []>("op_864_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_121_cast = mul(x = var_863_cast, y = var_864_to_fp16)[name = tensor<string, []>("aw_chunk_121_cast")];
            tensor<string, []> var_867_equation_0 = const()[name = tensor<string, []>("op_867_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_867_cast = einsum(equation = var_867_equation_0, values = (var_753_cast, var_722_cast))[name = tensor<string, []>("op_867_cast")];
            tensor<fp16, []> var_868_to_fp16 = const()[name = tensor<string, []>("op_868_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_123_cast = mul(x = var_867_cast, y = var_868_to_fp16)[name = tensor<string, []>("aw_chunk_123_cast")];
            tensor<string, []> var_871_equation_0 = const()[name = tensor<string, []>("op_871_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_871_cast = einsum(equation = var_871_equation_0, values = (var_753_cast, var_723_cast))[name = tensor<string, []>("op_871_cast")];
            tensor<fp16, []> var_872_to_fp16 = const()[name = tensor<string, []>("op_872_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_125_cast = mul(x = var_871_cast, y = var_872_to_fp16)[name = tensor<string, []>("aw_chunk_125_cast")];
            tensor<string, []> var_875_equation_0 = const()[name = tensor<string, []>("op_875_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_875_cast = einsum(equation = var_875_equation_0, values = (var_753_cast, var_724_cast))[name = tensor<string, []>("op_875_cast")];
            tensor<fp16, []> var_876_to_fp16 = const()[name = tensor<string, []>("op_876_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_127_cast = mul(x = var_875_cast, y = var_876_to_fp16)[name = tensor<string, []>("aw_chunk_127_cast")];
            tensor<string, []> var_879_equation_0 = const()[name = tensor<string, []>("op_879_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_879_cast = einsum(equation = var_879_equation_0, values = (var_757_cast, var_725_cast))[name = tensor<string, []>("op_879_cast")];
            tensor<fp16, []> var_880_to_fp16 = const()[name = tensor<string, []>("op_880_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_129_cast = mul(x = var_879_cast, y = var_880_to_fp16)[name = tensor<string, []>("aw_chunk_129_cast")];
            tensor<string, []> var_883_equation_0 = const()[name = tensor<string, []>("op_883_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_883_cast = einsum(equation = var_883_equation_0, values = (var_757_cast, var_726_cast))[name = tensor<string, []>("op_883_cast")];
            tensor<fp16, []> var_884_to_fp16 = const()[name = tensor<string, []>("op_884_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_131_cast = mul(x = var_883_cast, y = var_884_to_fp16)[name = tensor<string, []>("aw_chunk_131_cast")];
            tensor<string, []> var_887_equation_0 = const()[name = tensor<string, []>("op_887_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_887_cast = einsum(equation = var_887_equation_0, values = (var_757_cast, var_727_cast))[name = tensor<string, []>("op_887_cast")];
            tensor<fp16, []> var_888_to_fp16 = const()[name = tensor<string, []>("op_888_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_133_cast = mul(x = var_887_cast, y = var_888_to_fp16)[name = tensor<string, []>("aw_chunk_133_cast")];
            tensor<string, []> var_891_equation_0 = const()[name = tensor<string, []>("op_891_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_891_cast = einsum(equation = var_891_equation_0, values = (var_757_cast, var_728_cast))[name = tensor<string, []>("op_891_cast")];
            tensor<fp16, []> var_892_to_fp16 = const()[name = tensor<string, []>("op_892_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_135_cast = mul(x = var_891_cast, y = var_892_to_fp16)[name = tensor<string, []>("aw_chunk_135_cast")];
            tensor<string, []> var_895_equation_0 = const()[name = tensor<string, []>("op_895_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_895_cast = einsum(equation = var_895_equation_0, values = (var_757_cast, var_729_cast))[name = tensor<string, []>("op_895_cast")];
            tensor<fp16, []> var_896_to_fp16 = const()[name = tensor<string, []>("op_896_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_137_cast = mul(x = var_895_cast, y = var_896_to_fp16)[name = tensor<string, []>("aw_chunk_137_cast")];
            tensor<string, []> var_899_equation_0 = const()[name = tensor<string, []>("op_899_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_899_cast = einsum(equation = var_899_equation_0, values = (var_757_cast, var_730_cast))[name = tensor<string, []>("op_899_cast")];
            tensor<fp16, []> var_900_to_fp16 = const()[name = tensor<string, []>("op_900_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_139_cast = mul(x = var_899_cast, y = var_900_to_fp16)[name = tensor<string, []>("aw_chunk_139_cast")];
            tensor<string, []> var_903_equation_0 = const()[name = tensor<string, []>("op_903_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_903_cast = einsum(equation = var_903_equation_0, values = (var_757_cast, var_731_cast))[name = tensor<string, []>("op_903_cast")];
            tensor<fp16, []> var_904_to_fp16 = const()[name = tensor<string, []>("op_904_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_141_cast = mul(x = var_903_cast, y = var_904_to_fp16)[name = tensor<string, []>("aw_chunk_141_cast")];
            tensor<string, []> var_907_equation_0 = const()[name = tensor<string, []>("op_907_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_907_cast = einsum(equation = var_907_equation_0, values = (var_757_cast, var_732_cast))[name = tensor<string, []>("op_907_cast")];
            tensor<fp16, []> var_908_to_fp16 = const()[name = tensor<string, []>("op_908_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_143_cast = mul(x = var_907_cast, y = var_908_to_fp16)[name = tensor<string, []>("aw_chunk_143_cast")];
            tensor<string, []> var_911_equation_0 = const()[name = tensor<string, []>("op_911_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_911_cast = einsum(equation = var_911_equation_0, values = (var_761_cast, var_733_cast))[name = tensor<string, []>("op_911_cast")];
            tensor<fp16, []> var_912_to_fp16 = const()[name = tensor<string, []>("op_912_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_145_cast = mul(x = var_911_cast, y = var_912_to_fp16)[name = tensor<string, []>("aw_chunk_145_cast")];
            tensor<string, []> var_915_equation_0 = const()[name = tensor<string, []>("op_915_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_915_cast = einsum(equation = var_915_equation_0, values = (var_761_cast, var_734_cast))[name = tensor<string, []>("op_915_cast")];
            tensor<fp16, []> var_916_to_fp16 = const()[name = tensor<string, []>("op_916_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_147_cast = mul(x = var_915_cast, y = var_916_to_fp16)[name = tensor<string, []>("aw_chunk_147_cast")];
            tensor<string, []> var_919_equation_0 = const()[name = tensor<string, []>("op_919_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_919_cast = einsum(equation = var_919_equation_0, values = (var_761_cast, var_735_cast))[name = tensor<string, []>("op_919_cast")];
            tensor<fp16, []> var_920_to_fp16 = const()[name = tensor<string, []>("op_920_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_149_cast = mul(x = var_919_cast, y = var_920_to_fp16)[name = tensor<string, []>("aw_chunk_149_cast")];
            tensor<string, []> var_923_equation_0 = const()[name = tensor<string, []>("op_923_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_923_cast = einsum(equation = var_923_equation_0, values = (var_761_cast, var_736_cast))[name = tensor<string, []>("op_923_cast")];
            tensor<fp16, []> var_924_to_fp16 = const()[name = tensor<string, []>("op_924_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_151_cast = mul(x = var_923_cast, y = var_924_to_fp16)[name = tensor<string, []>("aw_chunk_151_cast")];
            tensor<string, []> var_927_equation_0 = const()[name = tensor<string, []>("op_927_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_927_cast = einsum(equation = var_927_equation_0, values = (var_761_cast, var_737_cast))[name = tensor<string, []>("op_927_cast")];
            tensor<fp16, []> var_928_to_fp16 = const()[name = tensor<string, []>("op_928_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_153_cast = mul(x = var_927_cast, y = var_928_to_fp16)[name = tensor<string, []>("aw_chunk_153_cast")];
            tensor<string, []> var_931_equation_0 = const()[name = tensor<string, []>("op_931_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_931_cast = einsum(equation = var_931_equation_0, values = (var_761_cast, var_738_cast))[name = tensor<string, []>("op_931_cast")];
            tensor<fp16, []> var_932_to_fp16 = const()[name = tensor<string, []>("op_932_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_155_cast = mul(x = var_931_cast, y = var_932_to_fp16)[name = tensor<string, []>("aw_chunk_155_cast")];
            tensor<string, []> var_935_equation_0 = const()[name = tensor<string, []>("op_935_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_935_cast = einsum(equation = var_935_equation_0, values = (var_761_cast, var_739_cast))[name = tensor<string, []>("op_935_cast")];
            tensor<fp16, []> var_936_to_fp16 = const()[name = tensor<string, []>("op_936_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_157_cast = mul(x = var_935_cast, y = var_936_to_fp16)[name = tensor<string, []>("aw_chunk_157_cast")];
            tensor<string, []> var_939_equation_0 = const()[name = tensor<string, []>("op_939_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_939_cast = einsum(equation = var_939_equation_0, values = (var_761_cast, var_740_cast))[name = tensor<string, []>("op_939_cast")];
            tensor<fp16, []> var_940_to_fp16 = const()[name = tensor<string, []>("op_940_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_159_cast = mul(x = var_939_cast, y = var_940_to_fp16)[name = tensor<string, []>("aw_chunk_159_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_942_cast = softmax(axis = var_120, x = aw_chunk_81_cast)[name = tensor<string, []>("op_942_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_943_cast = softmax(axis = var_120, x = aw_chunk_83_cast)[name = tensor<string, []>("op_943_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_944_cast = softmax(axis = var_120, x = aw_chunk_85_cast)[name = tensor<string, []>("op_944_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_945_cast = softmax(axis = var_120, x = aw_chunk_87_cast)[name = tensor<string, []>("op_945_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_946_cast = softmax(axis = var_120, x = aw_chunk_89_cast)[name = tensor<string, []>("op_946_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_947_cast = softmax(axis = var_120, x = aw_chunk_91_cast)[name = tensor<string, []>("op_947_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_948_cast = softmax(axis = var_120, x = aw_chunk_93_cast)[name = tensor<string, []>("op_948_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_949_cast = softmax(axis = var_120, x = aw_chunk_95_cast)[name = tensor<string, []>("op_949_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_950_cast = softmax(axis = var_120, x = aw_chunk_97_cast)[name = tensor<string, []>("op_950_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_951_cast = softmax(axis = var_120, x = aw_chunk_99_cast)[name = tensor<string, []>("op_951_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_952_cast = softmax(axis = var_120, x = aw_chunk_101_cast)[name = tensor<string, []>("op_952_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_953_cast = softmax(axis = var_120, x = aw_chunk_103_cast)[name = tensor<string, []>("op_953_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_954_cast = softmax(axis = var_120, x = aw_chunk_105_cast)[name = tensor<string, []>("op_954_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_955_cast = softmax(axis = var_120, x = aw_chunk_107_cast)[name = tensor<string, []>("op_955_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_956_cast = softmax(axis = var_120, x = aw_chunk_109_cast)[name = tensor<string, []>("op_956_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_957_cast = softmax(axis = var_120, x = aw_chunk_111_cast)[name = tensor<string, []>("op_957_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_958_cast = softmax(axis = var_120, x = aw_chunk_113_cast)[name = tensor<string, []>("op_958_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_959_cast = softmax(axis = var_120, x = aw_chunk_115_cast)[name = tensor<string, []>("op_959_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_960_cast = softmax(axis = var_120, x = aw_chunk_117_cast)[name = tensor<string, []>("op_960_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_961_cast = softmax(axis = var_120, x = aw_chunk_119_cast)[name = tensor<string, []>("op_961_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_962_cast = softmax(axis = var_120, x = aw_chunk_121_cast)[name = tensor<string, []>("op_962_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_963_cast = softmax(axis = var_120, x = aw_chunk_123_cast)[name = tensor<string, []>("op_963_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_964_cast = softmax(axis = var_120, x = aw_chunk_125_cast)[name = tensor<string, []>("op_964_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_965_cast = softmax(axis = var_120, x = aw_chunk_127_cast)[name = tensor<string, []>("op_965_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_966_cast = softmax(axis = var_120, x = aw_chunk_129_cast)[name = tensor<string, []>("op_966_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_967_cast = softmax(axis = var_120, x = aw_chunk_131_cast)[name = tensor<string, []>("op_967_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_968_cast = softmax(axis = var_120, x = aw_chunk_133_cast)[name = tensor<string, []>("op_968_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_969_cast = softmax(axis = var_120, x = aw_chunk_135_cast)[name = tensor<string, []>("op_969_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_970_cast = softmax(axis = var_120, x = aw_chunk_137_cast)[name = tensor<string, []>("op_970_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_971_cast = softmax(axis = var_120, x = aw_chunk_139_cast)[name = tensor<string, []>("op_971_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_972_cast = softmax(axis = var_120, x = aw_chunk_141_cast)[name = tensor<string, []>("op_972_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_973_cast = softmax(axis = var_120, x = aw_chunk_143_cast)[name = tensor<string, []>("op_973_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_974_cast = softmax(axis = var_120, x = aw_chunk_145_cast)[name = tensor<string, []>("op_974_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_975_cast = softmax(axis = var_120, x = aw_chunk_147_cast)[name = tensor<string, []>("op_975_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_976_cast = softmax(axis = var_120, x = aw_chunk_149_cast)[name = tensor<string, []>("op_976_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_977_cast = softmax(axis = var_120, x = aw_chunk_151_cast)[name = tensor<string, []>("op_977_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_978_cast = softmax(axis = var_120, x = aw_chunk_153_cast)[name = tensor<string, []>("op_978_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_979_cast = softmax(axis = var_120, x = aw_chunk_155_cast)[name = tensor<string, []>("op_979_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_980_cast = softmax(axis = var_120, x = aw_chunk_157_cast)[name = tensor<string, []>("op_980_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_981_cast = softmax(axis = var_120, x = aw_chunk_159_cast)[name = tensor<string, []>("op_981_cast")];
            tensor<string, []> var_983_equation_0 = const()[name = tensor<string, []>("op_983_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_983_cast = einsum(equation = var_983_equation_0, values = (var_763_cast, var_942_cast))[name = tensor<string, []>("op_983_cast")];
            tensor<string, []> var_985_equation_0 = const()[name = tensor<string, []>("op_985_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_985_cast = einsum(equation = var_985_equation_0, values = (var_763_cast, var_943_cast))[name = tensor<string, []>("op_985_cast")];
            tensor<string, []> var_987_equation_0 = const()[name = tensor<string, []>("op_987_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_987_cast = einsum(equation = var_987_equation_0, values = (var_763_cast, var_944_cast))[name = tensor<string, []>("op_987_cast")];
            tensor<string, []> var_989_equation_0 = const()[name = tensor<string, []>("op_989_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_989_cast = einsum(equation = var_989_equation_0, values = (var_763_cast, var_945_cast))[name = tensor<string, []>("op_989_cast")];
            tensor<string, []> var_991_equation_0 = const()[name = tensor<string, []>("op_991_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_991_cast = einsum(equation = var_991_equation_0, values = (var_763_cast, var_946_cast))[name = tensor<string, []>("op_991_cast")];
            tensor<string, []> var_993_equation_0 = const()[name = tensor<string, []>("op_993_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_993_cast = einsum(equation = var_993_equation_0, values = (var_763_cast, var_947_cast))[name = tensor<string, []>("op_993_cast")];
            tensor<string, []> var_995_equation_0 = const()[name = tensor<string, []>("op_995_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_995_cast = einsum(equation = var_995_equation_0, values = (var_763_cast, var_948_cast))[name = tensor<string, []>("op_995_cast")];
            tensor<string, []> var_997_equation_0 = const()[name = tensor<string, []>("op_997_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_997_cast = einsum(equation = var_997_equation_0, values = (var_763_cast, var_949_cast))[name = tensor<string, []>("op_997_cast")];
            tensor<string, []> var_999_equation_0 = const()[name = tensor<string, []>("op_999_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_999_cast = einsum(equation = var_999_equation_0, values = (var_767_cast, var_950_cast))[name = tensor<string, []>("op_999_cast")];
            tensor<string, []> var_1001_equation_0 = const()[name = tensor<string, []>("op_1001_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1001_cast = einsum(equation = var_1001_equation_0, values = (var_767_cast, var_951_cast))[name = tensor<string, []>("op_1001_cast")];
            tensor<string, []> var_1003_equation_0 = const()[name = tensor<string, []>("op_1003_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1003_cast = einsum(equation = var_1003_equation_0, values = (var_767_cast, var_952_cast))[name = tensor<string, []>("op_1003_cast")];
            tensor<string, []> var_1005_equation_0 = const()[name = tensor<string, []>("op_1005_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1005_cast = einsum(equation = var_1005_equation_0, values = (var_767_cast, var_953_cast))[name = tensor<string, []>("op_1005_cast")];
            tensor<string, []> var_1007_equation_0 = const()[name = tensor<string, []>("op_1007_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1007_cast = einsum(equation = var_1007_equation_0, values = (var_767_cast, var_954_cast))[name = tensor<string, []>("op_1007_cast")];
            tensor<string, []> var_1009_equation_0 = const()[name = tensor<string, []>("op_1009_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1009_cast = einsum(equation = var_1009_equation_0, values = (var_767_cast, var_955_cast))[name = tensor<string, []>("op_1009_cast")];
            tensor<string, []> var_1011_equation_0 = const()[name = tensor<string, []>("op_1011_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1011_cast = einsum(equation = var_1011_equation_0, values = (var_767_cast, var_956_cast))[name = tensor<string, []>("op_1011_cast")];
            tensor<string, []> var_1013_equation_0 = const()[name = tensor<string, []>("op_1013_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1013_cast = einsum(equation = var_1013_equation_0, values = (var_767_cast, var_957_cast))[name = tensor<string, []>("op_1013_cast")];
            tensor<string, []> var_1015_equation_0 = const()[name = tensor<string, []>("op_1015_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1015_cast = einsum(equation = var_1015_equation_0, values = (var_771_cast, var_958_cast))[name = tensor<string, []>("op_1015_cast")];
            tensor<string, []> var_1017_equation_0 = const()[name = tensor<string, []>("op_1017_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1017_cast = einsum(equation = var_1017_equation_0, values = (var_771_cast, var_959_cast))[name = tensor<string, []>("op_1017_cast")];
            tensor<string, []> var_1019_equation_0 = const()[name = tensor<string, []>("op_1019_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1019_cast = einsum(equation = var_1019_equation_0, values = (var_771_cast, var_960_cast))[name = tensor<string, []>("op_1019_cast")];
            tensor<string, []> var_1021_equation_0 = const()[name = tensor<string, []>("op_1021_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1021_cast = einsum(equation = var_1021_equation_0, values = (var_771_cast, var_961_cast))[name = tensor<string, []>("op_1021_cast")];
            tensor<string, []> var_1023_equation_0 = const()[name = tensor<string, []>("op_1023_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1023_cast = einsum(equation = var_1023_equation_0, values = (var_771_cast, var_962_cast))[name = tensor<string, []>("op_1023_cast")];
            tensor<string, []> var_1025_equation_0 = const()[name = tensor<string, []>("op_1025_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1025_cast = einsum(equation = var_1025_equation_0, values = (var_771_cast, var_963_cast))[name = tensor<string, []>("op_1025_cast")];
            tensor<string, []> var_1027_equation_0 = const()[name = tensor<string, []>("op_1027_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1027_cast = einsum(equation = var_1027_equation_0, values = (var_771_cast, var_964_cast))[name = tensor<string, []>("op_1027_cast")];
            tensor<string, []> var_1029_equation_0 = const()[name = tensor<string, []>("op_1029_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1029_cast = einsum(equation = var_1029_equation_0, values = (var_771_cast, var_965_cast))[name = tensor<string, []>("op_1029_cast")];
            tensor<string, []> var_1031_equation_0 = const()[name = tensor<string, []>("op_1031_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1031_cast = einsum(equation = var_1031_equation_0, values = (var_775_cast, var_966_cast))[name = tensor<string, []>("op_1031_cast")];
            tensor<string, []> var_1033_equation_0 = const()[name = tensor<string, []>("op_1033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1033_cast = einsum(equation = var_1033_equation_0, values = (var_775_cast, var_967_cast))[name = tensor<string, []>("op_1033_cast")];
            tensor<string, []> var_1035_equation_0 = const()[name = tensor<string, []>("op_1035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1035_cast = einsum(equation = var_1035_equation_0, values = (var_775_cast, var_968_cast))[name = tensor<string, []>("op_1035_cast")];
            tensor<string, []> var_1037_equation_0 = const()[name = tensor<string, []>("op_1037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1037_cast = einsum(equation = var_1037_equation_0, values = (var_775_cast, var_969_cast))[name = tensor<string, []>("op_1037_cast")];
            tensor<string, []> var_1039_equation_0 = const()[name = tensor<string, []>("op_1039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1039_cast = einsum(equation = var_1039_equation_0, values = (var_775_cast, var_970_cast))[name = tensor<string, []>("op_1039_cast")];
            tensor<string, []> var_1041_equation_0 = const()[name = tensor<string, []>("op_1041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1041_cast = einsum(equation = var_1041_equation_0, values = (var_775_cast, var_971_cast))[name = tensor<string, []>("op_1041_cast")];
            tensor<string, []> var_1043_equation_0 = const()[name = tensor<string, []>("op_1043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1043_cast = einsum(equation = var_1043_equation_0, values = (var_775_cast, var_972_cast))[name = tensor<string, []>("op_1043_cast")];
            tensor<string, []> var_1045_equation_0 = const()[name = tensor<string, []>("op_1045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1045_cast = einsum(equation = var_1045_equation_0, values = (var_775_cast, var_973_cast))[name = tensor<string, []>("op_1045_cast")];
            tensor<string, []> var_1047_equation_0 = const()[name = tensor<string, []>("op_1047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1047_cast = einsum(equation = var_1047_equation_0, values = (var_779_cast, var_974_cast))[name = tensor<string, []>("op_1047_cast")];
            tensor<string, []> var_1049_equation_0 = const()[name = tensor<string, []>("op_1049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1049_cast = einsum(equation = var_1049_equation_0, values = (var_779_cast, var_975_cast))[name = tensor<string, []>("op_1049_cast")];
            tensor<string, []> var_1051_equation_0 = const()[name = tensor<string, []>("op_1051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1051_cast = einsum(equation = var_1051_equation_0, values = (var_779_cast, var_976_cast))[name = tensor<string, []>("op_1051_cast")];
            tensor<string, []> var_1053_equation_0 = const()[name = tensor<string, []>("op_1053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1053_cast = einsum(equation = var_1053_equation_0, values = (var_779_cast, var_977_cast))[name = tensor<string, []>("op_1053_cast")];
            tensor<string, []> var_1055_equation_0 = const()[name = tensor<string, []>("op_1055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1055_cast = einsum(equation = var_1055_equation_0, values = (var_779_cast, var_978_cast))[name = tensor<string, []>("op_1055_cast")];
            tensor<string, []> var_1057_equation_0 = const()[name = tensor<string, []>("op_1057_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1057_cast = einsum(equation = var_1057_equation_0, values = (var_779_cast, var_979_cast))[name = tensor<string, []>("op_1057_cast")];
            tensor<string, []> var_1059_equation_0 = const()[name = tensor<string, []>("op_1059_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1059_cast = einsum(equation = var_1059_equation_0, values = (var_779_cast, var_980_cast))[name = tensor<string, []>("op_1059_cast")];
            tensor<string, []> var_1061_equation_0 = const()[name = tensor<string, []>("op_1061_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1061_cast = einsum(equation = var_1061_equation_0, values = (var_779_cast, var_981_cast))[name = tensor<string, []>("op_1061_cast")];
            tensor<bool, []> var_1063_interleave_0 = const()[name = tensor<string, []>("op_1063_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1063_cast = concat(axis = var_95, interleave = var_1063_interleave_0, values = (var_983_cast, var_985_cast, var_987_cast, var_989_cast, var_991_cast, var_993_cast, var_995_cast, var_997_cast))[name = tensor<string, []>("op_1063_cast")];
            tensor<bool, []> var_1065_interleave_0 = const()[name = tensor<string, []>("op_1065_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1065_cast = concat(axis = var_95, interleave = var_1065_interleave_0, values = (var_999_cast, var_1001_cast, var_1003_cast, var_1005_cast, var_1007_cast, var_1009_cast, var_1011_cast, var_1013_cast))[name = tensor<string, []>("op_1065_cast")];
            tensor<bool, []> var_1067_interleave_0 = const()[name = tensor<string, []>("op_1067_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1067_cast = concat(axis = var_95, interleave = var_1067_interleave_0, values = (var_1015_cast, var_1017_cast, var_1019_cast, var_1021_cast, var_1023_cast, var_1025_cast, var_1027_cast, var_1029_cast))[name = tensor<string, []>("op_1067_cast")];
            tensor<bool, []> var_1069_interleave_0 = const()[name = tensor<string, []>("op_1069_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1069_cast = concat(axis = var_95, interleave = var_1069_interleave_0, values = (var_1031_cast, var_1033_cast, var_1035_cast, var_1037_cast, var_1039_cast, var_1041_cast, var_1043_cast, var_1045_cast))[name = tensor<string, []>("op_1069_cast")];
            tensor<bool, []> var_1071_interleave_0 = const()[name = tensor<string, []>("op_1071_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1071_cast = concat(axis = var_95, interleave = var_1071_interleave_0, values = (var_1047_cast, var_1049_cast, var_1051_cast, var_1053_cast, var_1055_cast, var_1057_cast, var_1059_cast, var_1061_cast))[name = tensor<string, []>("op_1071_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_120, interleave = input_27_interleave_0, values = (var_1063_cast, var_1065_cast, var_1067_cast, var_1069_cast, var_1071_cast))[name = tensor<string, []>("input_27_cast")];
            tensor<int32, [2]> var_1077 = const()[name = tensor<string, []>("op_1077"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1079 = const()[name = tensor<string, []>("op_1079"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1081_pad_type_0 = const()[name = tensor<string, []>("op_1081_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1081_pad_0 = const()[name = tensor<string, []>("op_1081_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414784))), lut = tensor<fp16, [4]>([-0x1.33cp-6, -0x1.9f4p-9, 0x1.a58p-9, 0x1.348p-6]), 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, []>(1440448)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1081_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_1079, groups = var_120, pad = var_1081_pad_0, pad_type = var_1081_pad_type_0, strides = var_1077, 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_1081_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_5_cast = add(x = var_1081_cast, y = inputs_3_cast)[name = tensor<string, []>("inputs_5_cast")];
            tensor<int32, [1]> var_1085 = const()[name = tensor<string, []>("op_1085"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_5_cast = reduce_mean(axes = var_1085, keep_dims = var_115, 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_1089 = const()[name = tensor<string, []>("op_1089"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1090_cast = reduce_mean(axes = var_1089, keep_dims = var_115, x = zero_mean_sq_5_cast)[name = tensor<string, []>("op_1090_cast")];
            tensor<fp16, []> var_1091_to_fp16 = const()[name = tensor<string, []>("op_1091_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1092_cast = add(x = var_1090_cast, y = var_1091_to_fp16)[name = tensor<string, []>("op_1092_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_1092_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_1096_to_fp16 = const()[name = tensor<string, []>("op_1096_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1441152)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1097_cast = add(x = out_5_cast, y = var_1096_to_fp16)[name = tensor<string, []>("op_1097_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1099_to_fp16 = const()[name = tensor<string, []>("op_1099_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1441856)))];
            tensor<fp16, [2, 320, 1, 4096]> input_29_cast = mul(x = var_1097_cast, y = var_1099_to_fp16)[name = tensor<string, []>("input_29_cast")];
            tensor<int32, [2]> var_1107 = const()[name = tensor<string, []>("op_1107"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1109 = const()[name = tensor<string, []>("op_1109"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1111_pad_type_0 = const()[name = tensor<string, []>("op_1111_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1111_pad_0 = const()[name = tensor<string, []>("op_1111_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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1442560))), lut = tensor<fp16, [4]>([-0x1.9fp-4, -0x1.de8p-6, 0x1.e04p-6, 0x1.9ecp-4]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1647424))), lut = tensor<fp16, [4]>([-0x1.52p-6, -0x1.bf4p-3, 0x1.d7cp-6, -0x1.5fp-4]), 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_1111_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_1109, groups = var_120, pad = var_1111_pad_0, pad_type = var_1111_pad_type_0, strides = var_1107, 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_1111_cast")];
            tensor<int32, [2]> var_1112_split_sizes_0 = const()[name = tensor<string, []>("op_1112_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_1112_axis_0 = const()[name = tensor<string, []>("op_1112_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_1112_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_1112_cast_1 = split(axis = var_1112_axis_0, split_sizes = var_1112_split_sizes_0, x = var_1111_cast)[name = tensor<string, []>("op_1112_cast")];
            tensor<string, []> var_1114_mode_0 = const()[name = tensor<string, []>("op_1114_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_1114_cast = gelu(mode = var_1114_mode_0, x = var_1112_cast_1)[name = tensor<string, []>("op_1114_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_31_cast = mul(x = var_1112_cast_0, y = var_1114_cast)[name = tensor<string, []>("input_31_cast")];
            tensor<int32, [2]> var_1118 = const()[name = tensor<string, []>("op_1118"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1120 = const()[name = tensor<string, []>("op_1120"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1122_pad_type_0 = const()[name = tensor<string, []>("op_1122_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1122_pad_0 = const()[name = tensor<string, []>("op_1122_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1648128))), lut = tensor<fp16, [4]>([-0x1.748p-4, -0x1.b48p-6, 0x1.c4cp-6, 0x1.77cp-4]), 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, []>(1750592)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1122_cast = conv(bias = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_1120, groups = var_120, pad = var_1122_pad_0, pad_type = var_1122_pad_type_0, strides = var_1118, weight = down_blocks_0_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_31_cast)[name = tensor<string, []>("op_1122_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_15_cast = add(x = var_1122_cast, y = inputs_5_cast)[name = tensor<string, []>("hidden_states_15_cast")];
            tensor<int32, [4]> var_1124 = const()[name = tensor<string, []>("op_1124"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_33_cast = reshape(shape = var_1124, x = hidden_states_15_cast)[name = tensor<string, []>("input_33_cast")];
            tensor<int32, [2]> var_1128 = const()[name = tensor<string, []>("op_1128"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1130 = const()[name = tensor<string, []>("op_1130"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1751296))), lut = tensor<fp16, [4]>([-0x1.394p-4, -0x1.74p-6, 0x1.76p-6, 0x1.3bcp-4]), 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, []>(1776960)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_17_cast = conv(bias = down_blocks_0_attentions_0_proj_out_bias_to_fp16, dilations = var_1130, groups = var_120, pad = hidden_states_17_pad_0, pad_type = hidden_states_17_pad_type_0, strides = var_1128, 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, []>(1777664)))];
            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, []>(1778368)))];
            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_1145 = const()[name = tensor<string, []>("op_1145"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1147 = const()[name = tensor<string, []>("op_1147"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1779072))), lut = tensor<fp16, [4]>([-0x1.a5p-4, -0x1.b48p-6, 0x1.a5cp-6, 0x1.9f4p-4]), 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, []>(2009536)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_19_cast = conv(bias = down_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_1147, groups = var_120, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = var_1145, 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_1153 = const()[name = tensor<string, []>("op_1153"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1155 = const()[name = tensor<string, []>("op_1155"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2010240))), lut = tensor<fp16, [4]>([-0x1.6b4p-5, -0x1.c28p-9, 0x1.aa4p-9, 0x1.184p-5]), 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, []>(2112704)))];
            tensor<fp16, [2, 320, 1, 1]> temb_3_cast = conv(bias = down_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_1155, groups = var_120, pad = temb_3_pad_0, pad_type = temb_3_pad_type_0, strides = var_1153, 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, []>(2113408)))];
            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, []>(2114112)))];
            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_1165 = const()[name = tensor<string, []>("op_1165"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1167 = const()[name = tensor<string, []>("op_1167"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2114816))), lut = tensor<fp16, [4]>([-0x1.648p-4, -0x1.724p-6, 0x1.554p-6, 0x1.594p-4]), 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, []>(2345280)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_21_cast = conv(bias = down_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_1167, groups = var_120, pad = hidden_states_21_pad_0, pad_type = hidden_states_21_pad_type_0, strides = var_1165, 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, []>(2345984)))];
            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, []>(2346688)))];
            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_1187 = const()[name = tensor<string, []>("op_1187"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1189 = const()[name = tensor<string, []>("op_1189"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2347392))), lut = tensor<fp16, [4]>([-0x1.788p-4, -0x1.bdcp-6, 0x1.b58p-6, 0x1.74p-4]), 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, []>(2373056)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_25_cast = conv(bias = down_blocks_0_attentions_1_proj_in_bias_to_fp16, dilations = var_1189, groups = var_120, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = var_1187, 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_1194 = const()[name = tensor<string, []>("op_1194"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_7_cast = reshape(shape = var_1194, x = hidden_states_25_cast)[name = tensor<string, []>("inputs_7_cast")];
            tensor<int32, [1]> var_1204 = const()[name = tensor<string, []>("op_1204"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_7_cast = reduce_mean(axes = var_1204, keep_dims = var_115, 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_1208 = const()[name = tensor<string, []>("op_1208"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1209_cast = reduce_mean(axes = var_1208, keep_dims = var_115, x = zero_mean_sq_7_cast)[name = tensor<string, []>("op_1209_cast")];
            tensor<fp16, []> var_1210_to_fp16 = const()[name = tensor<string, []>("op_1210_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1211_cast = add(x = var_1209_cast, y = var_1210_to_fp16)[name = tensor<string, []>("op_1211_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_1211_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_1215_to_fp16 = const()[name = tensor<string, []>("op_1215_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2373760)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1216_cast = add(x = out_7_cast, y = var_1215_to_fp16)[name = tensor<string, []>("op_1216_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1218_to_fp16 = const()[name = tensor<string, []>("op_1218_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2374464)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_27_cast = mul(x = var_1216_cast, y = var_1218_to_fp16)[name = tensor<string, []>("hidden_states_27_cast")];
            tensor<int32, [2]> var_1225 = const()[name = tensor<string, []>("op_1225"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1227 = const()[name = tensor<string, []>("op_1227"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2375168))), lut = tensor<fp16, [4]>([-0x1.3dp-3, -0x1.7e8p-5, 0x1.704p-5, 0x1.394p-3]), 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_1227, groups = var_120, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = var_1225, 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_1231 = const()[name = tensor<string, []>("op_1231"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1233 = const()[name = tensor<string, []>("op_1233"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2400832))), lut = tensor<fp16, [4]>([-0x1.234p-3, -0x1.57p-5, 0x1.54p-5, 0x1.208p-3]), 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_1233, groups = var_120, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = var_1231, 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_1237 = const()[name = tensor<string, []>("op_1237"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1239 = const()[name = tensor<string, []>("op_1239"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2426496))), lut = tensor<fp16, [4]>([-0x1.624p-4, -0x1.97p-6, 0x1.994p-6, 0x1.608p-4]), 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_1239, groups = var_120, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = var_1237, 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_1243_begin_0 = const()[name = tensor<string, []>("op_1243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1243_end_0 = const()[name = tensor<string, []>("op_1243_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1243_end_mask_0 = const()[name = tensor<string, []>("op_1243_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1243_cast = slice_by_index(begin = var_1243_begin_0, end = var_1243_end_0, end_mask = var_1243_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1243_cast")];
            tensor<int32, [4]> var_1247_begin_0 = const()[name = tensor<string, []>("op_1247_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1247_end_0 = const()[name = tensor<string, []>("op_1247_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1247_end_mask_0 = const()[name = tensor<string, []>("op_1247_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1247_cast = slice_by_index(begin = var_1247_begin_0, end = var_1247_end_0, end_mask = var_1247_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1247_cast")];
            tensor<int32, [4]> var_1251_begin_0 = const()[name = tensor<string, []>("op_1251_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1251_end_0 = const()[name = tensor<string, []>("op_1251_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1251_end_mask_0 = const()[name = tensor<string, []>("op_1251_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1251_cast = slice_by_index(begin = var_1251_begin_0, end = var_1251_end_0, end_mask = var_1251_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1251_cast")];
            tensor<int32, [4]> var_1255_begin_0 = const()[name = tensor<string, []>("op_1255_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1255_end_0 = const()[name = tensor<string, []>("op_1255_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1255_end_mask_0 = const()[name = tensor<string, []>("op_1255_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1255_cast = slice_by_index(begin = var_1255_begin_0, end = var_1255_end_0, end_mask = var_1255_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1255_cast")];
            tensor<int32, [4]> var_1259_begin_0 = const()[name = tensor<string, []>("op_1259_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1259_end_0 = const()[name = tensor<string, []>("op_1259_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1259_end_mask_0 = const()[name = tensor<string, []>("op_1259_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1259_cast = slice_by_index(begin = var_1259_begin_0, end = var_1259_end_0, end_mask = var_1259_end_mask_0, x = q_5_cast)[name = tensor<string, []>("op_1259_cast")];
            tensor<int32, [4]> var_1262_begin_0 = const()[name = tensor<string, []>("op_1262_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1262_end_0 = const()[name = tensor<string, []>("op_1262_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1262_end_mask_0 = const()[name = tensor<string, []>("op_1262_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1262_cast = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1262_cast")];
            tensor<int32, [4]> var_1263_begin_0 = const()[name = tensor<string, []>("op_1263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1263_end_0 = const()[name = tensor<string, []>("op_1263_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1263_end_mask_0 = const()[name = tensor<string, []>("op_1263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1263_cast = slice_by_index(begin = var_1263_begin_0, end = var_1263_end_0, end_mask = var_1263_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1263_cast")];
            tensor<int32, [4]> var_1264_begin_0 = const()[name = tensor<string, []>("op_1264_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1264_end_0 = const()[name = tensor<string, []>("op_1264_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1264_end_mask_0 = const()[name = tensor<string, []>("op_1264_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1264_cast = slice_by_index(begin = var_1264_begin_0, end = var_1264_end_0, end_mask = var_1264_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1264_cast")];
            tensor<int32, [4]> var_1265_begin_0 = const()[name = tensor<string, []>("op_1265_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1265_end_0 = const()[name = tensor<string, []>("op_1265_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1265_end_mask_0 = const()[name = tensor<string, []>("op_1265_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1265_cast = slice_by_index(begin = var_1265_begin_0, end = var_1265_end_0, end_mask = var_1265_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1265_cast")];
            tensor<int32, [4]> var_1266_begin_0 = const()[name = tensor<string, []>("op_1266_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1266_end_0 = const()[name = tensor<string, []>("op_1266_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1266_end_mask_0 = const()[name = tensor<string, []>("op_1266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1266_cast = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1266_cast")];
            tensor<int32, [4]> var_1267_begin_0 = const()[name = tensor<string, []>("op_1267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1267_end_0 = const()[name = tensor<string, []>("op_1267_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1267_end_mask_0 = const()[name = tensor<string, []>("op_1267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1267_cast = slice_by_index(begin = var_1267_begin_0, end = var_1267_end_0, end_mask = var_1267_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1267_cast")];
            tensor<int32, [4]> var_1268_begin_0 = const()[name = tensor<string, []>("op_1268_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1268_end_0 = const()[name = tensor<string, []>("op_1268_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1268_end_mask_0 = const()[name = tensor<string, []>("op_1268_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1268_cast = slice_by_index(begin = var_1268_begin_0, end = var_1268_end_0, end_mask = var_1268_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1268_cast")];
            tensor<int32, [4]> var_1269_begin_0 = const()[name = tensor<string, []>("op_1269_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1269_end_0 = const()[name = tensor<string, []>("op_1269_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1269_end_mask_0 = const()[name = tensor<string, []>("op_1269_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1269_cast = slice_by_index(begin = var_1269_begin_0, end = var_1269_end_0, end_mask = var_1269_end_mask_0, x = var_1243_cast)[name = tensor<string, []>("op_1269_cast")];
            tensor<int32, [4]> var_1270_begin_0 = const()[name = tensor<string, []>("op_1270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1270_end_0 = const()[name = tensor<string, []>("op_1270_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1270_end_mask_0 = const()[name = tensor<string, []>("op_1270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1270_cast = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1270_cast")];
            tensor<int32, [4]> var_1271_begin_0 = const()[name = tensor<string, []>("op_1271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1271_end_0 = const()[name = tensor<string, []>("op_1271_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1271_end_mask_0 = const()[name = tensor<string, []>("op_1271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1271_cast = slice_by_index(begin = var_1271_begin_0, end = var_1271_end_0, end_mask = var_1271_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1271_cast")];
            tensor<int32, [4]> var_1272_begin_0 = const()[name = tensor<string, []>("op_1272_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1272_end_0 = const()[name = tensor<string, []>("op_1272_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1272_end_mask_0 = const()[name = tensor<string, []>("op_1272_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1272_cast = slice_by_index(begin = var_1272_begin_0, end = var_1272_end_0, end_mask = var_1272_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1272_cast")];
            tensor<int32, [4]> var_1273_begin_0 = const()[name = tensor<string, []>("op_1273_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1273_end_0 = const()[name = tensor<string, []>("op_1273_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1273_end_mask_0 = const()[name = tensor<string, []>("op_1273_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1273_cast = slice_by_index(begin = var_1273_begin_0, end = var_1273_end_0, end_mask = var_1273_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1273_cast")];
            tensor<int32, [4]> var_1274_begin_0 = const()[name = tensor<string, []>("op_1274_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1274_end_0 = const()[name = tensor<string, []>("op_1274_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1274_end_mask_0 = const()[name = tensor<string, []>("op_1274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1274_cast = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1274_cast")];
            tensor<int32, [4]> var_1275_begin_0 = const()[name = tensor<string, []>("op_1275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1275_end_0 = const()[name = tensor<string, []>("op_1275_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1275_end_mask_0 = const()[name = tensor<string, []>("op_1275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1275_cast = slice_by_index(begin = var_1275_begin_0, end = var_1275_end_0, end_mask = var_1275_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1275_cast")];
            tensor<int32, [4]> var_1276_begin_0 = const()[name = tensor<string, []>("op_1276_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1276_end_0 = const()[name = tensor<string, []>("op_1276_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1276_end_mask_0 = const()[name = tensor<string, []>("op_1276_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1276_cast = slice_by_index(begin = var_1276_begin_0, end = var_1276_end_0, end_mask = var_1276_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1276_cast")];
            tensor<int32, [4]> var_1277_begin_0 = const()[name = tensor<string, []>("op_1277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1277_end_0 = const()[name = tensor<string, []>("op_1277_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1277_end_mask_0 = const()[name = tensor<string, []>("op_1277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1277_cast = slice_by_index(begin = var_1277_begin_0, end = var_1277_end_0, end_mask = var_1277_end_mask_0, x = var_1247_cast)[name = tensor<string, []>("op_1277_cast")];
            tensor<int32, [4]> var_1278_begin_0 = const()[name = tensor<string, []>("op_1278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1278_end_0 = const()[name = tensor<string, []>("op_1278_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1278_end_mask_0 = const()[name = tensor<string, []>("op_1278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1278_cast = slice_by_index(begin = var_1278_begin_0, end = var_1278_end_0, end_mask = var_1278_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1278_cast")];
            tensor<int32, [4]> var_1279_begin_0 = const()[name = tensor<string, []>("op_1279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1279_end_0 = const()[name = tensor<string, []>("op_1279_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1279_end_mask_0 = const()[name = tensor<string, []>("op_1279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1279_cast = slice_by_index(begin = var_1279_begin_0, end = var_1279_end_0, end_mask = var_1279_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1279_cast")];
            tensor<int32, [4]> var_1280_begin_0 = const()[name = tensor<string, []>("op_1280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1280_end_0 = const()[name = tensor<string, []>("op_1280_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1280_end_mask_0 = const()[name = tensor<string, []>("op_1280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1280_cast = slice_by_index(begin = var_1280_begin_0, end = var_1280_end_0, end_mask = var_1280_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1280_cast")];
            tensor<int32, [4]> var_1281_begin_0 = const()[name = tensor<string, []>("op_1281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1281_end_0 = const()[name = tensor<string, []>("op_1281_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1281_end_mask_0 = const()[name = tensor<string, []>("op_1281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1281_cast = slice_by_index(begin = var_1281_begin_0, end = var_1281_end_0, end_mask = var_1281_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1281_cast")];
            tensor<int32, [4]> var_1282_begin_0 = const()[name = tensor<string, []>("op_1282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1282_end_0 = const()[name = tensor<string, []>("op_1282_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1282_end_mask_0 = const()[name = tensor<string, []>("op_1282_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1282_cast = slice_by_index(begin = var_1282_begin_0, end = var_1282_end_0, end_mask = var_1282_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1282_cast")];
            tensor<int32, [4]> var_1283_begin_0 = const()[name = tensor<string, []>("op_1283_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1283_end_0 = const()[name = tensor<string, []>("op_1283_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1283_end_mask_0 = const()[name = tensor<string, []>("op_1283_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1283_cast = slice_by_index(begin = var_1283_begin_0, end = var_1283_end_0, end_mask = var_1283_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1283_cast")];
            tensor<int32, [4]> var_1284_begin_0 = const()[name = tensor<string, []>("op_1284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1284_end_0 = const()[name = tensor<string, []>("op_1284_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1284_end_mask_0 = const()[name = tensor<string, []>("op_1284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1284_cast = slice_by_index(begin = var_1284_begin_0, end = var_1284_end_0, end_mask = var_1284_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1284_cast")];
            tensor<int32, [4]> var_1285_begin_0 = const()[name = tensor<string, []>("op_1285_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1285_end_0 = const()[name = tensor<string, []>("op_1285_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1285_end_mask_0 = const()[name = tensor<string, []>("op_1285_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1285_cast = slice_by_index(begin = var_1285_begin_0, end = var_1285_end_0, end_mask = var_1285_end_mask_0, x = var_1251_cast)[name = tensor<string, []>("op_1285_cast")];
            tensor<int32, [4]> var_1286_begin_0 = const()[name = tensor<string, []>("op_1286_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1286_end_0 = const()[name = tensor<string, []>("op_1286_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1286_end_mask_0 = const()[name = tensor<string, []>("op_1286_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1286_cast = slice_by_index(begin = var_1286_begin_0, end = var_1286_end_0, end_mask = var_1286_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1286_cast")];
            tensor<int32, [4]> var_1287_begin_0 = const()[name = tensor<string, []>("op_1287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1287_end_0 = const()[name = tensor<string, []>("op_1287_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1287_end_mask_0 = const()[name = tensor<string, []>("op_1287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1287_cast = slice_by_index(begin = var_1287_begin_0, end = var_1287_end_0, end_mask = var_1287_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1287_cast")];
            tensor<int32, [4]> var_1288_begin_0 = const()[name = tensor<string, []>("op_1288_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1288_end_0 = const()[name = tensor<string, []>("op_1288_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1288_end_mask_0 = const()[name = tensor<string, []>("op_1288_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1288_cast = slice_by_index(begin = var_1288_begin_0, end = var_1288_end_0, end_mask = var_1288_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1288_cast")];
            tensor<int32, [4]> var_1289_begin_0 = const()[name = tensor<string, []>("op_1289_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1289_end_0 = const()[name = tensor<string, []>("op_1289_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1289_end_mask_0 = const()[name = tensor<string, []>("op_1289_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1289_cast = slice_by_index(begin = var_1289_begin_0, end = var_1289_end_0, end_mask = var_1289_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1289_cast")];
            tensor<int32, [4]> var_1290_begin_0 = const()[name = tensor<string, []>("op_1290_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1290_end_0 = const()[name = tensor<string, []>("op_1290_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1290_end_mask_0 = const()[name = tensor<string, []>("op_1290_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1290_cast = slice_by_index(begin = var_1290_begin_0, end = var_1290_end_0, end_mask = var_1290_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1290_cast")];
            tensor<int32, [4]> var_1291_begin_0 = const()[name = tensor<string, []>("op_1291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1291_end_0 = const()[name = tensor<string, []>("op_1291_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1291_end_mask_0 = const()[name = tensor<string, []>("op_1291_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1291_cast = slice_by_index(begin = var_1291_begin_0, end = var_1291_end_0, end_mask = var_1291_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1291_cast")];
            tensor<int32, [4]> var_1292_begin_0 = const()[name = tensor<string, []>("op_1292_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1292_end_0 = const()[name = tensor<string, []>("op_1292_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1292_end_mask_0 = const()[name = tensor<string, []>("op_1292_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1292_cast = slice_by_index(begin = var_1292_begin_0, end = var_1292_end_0, end_mask = var_1292_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1292_cast")];
            tensor<int32, [4]> var_1293_begin_0 = const()[name = tensor<string, []>("op_1293_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1293_end_0 = const()[name = tensor<string, []>("op_1293_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1293_end_mask_0 = const()[name = tensor<string, []>("op_1293_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1293_cast = slice_by_index(begin = var_1293_begin_0, end = var_1293_end_0, end_mask = var_1293_end_mask_0, x = var_1255_cast)[name = tensor<string, []>("op_1293_cast")];
            tensor<int32, [4]> var_1294_begin_0 = const()[name = tensor<string, []>("op_1294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1294_end_0 = const()[name = tensor<string, []>("op_1294_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1294_end_mask_0 = const()[name = tensor<string, []>("op_1294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1294_cast = slice_by_index(begin = var_1294_begin_0, end = var_1294_end_0, end_mask = var_1294_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1294_cast")];
            tensor<int32, [4]> var_1295_begin_0 = const()[name = tensor<string, []>("op_1295_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1295_end_0 = const()[name = tensor<string, []>("op_1295_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1295_end_mask_0 = const()[name = tensor<string, []>("op_1295_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1295_cast = slice_by_index(begin = var_1295_begin_0, end = var_1295_end_0, end_mask = var_1295_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1295_cast")];
            tensor<int32, [4]> var_1296_begin_0 = const()[name = tensor<string, []>("op_1296_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1296_end_0 = const()[name = tensor<string, []>("op_1296_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1296_end_mask_0 = const()[name = tensor<string, []>("op_1296_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1296_cast = slice_by_index(begin = var_1296_begin_0, end = var_1296_end_0, end_mask = var_1296_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1296_cast")];
            tensor<int32, [4]> var_1297_begin_0 = const()[name = tensor<string, []>("op_1297_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1297_end_0 = const()[name = tensor<string, []>("op_1297_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1297_end_mask_0 = const()[name = tensor<string, []>("op_1297_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1297_cast = slice_by_index(begin = var_1297_begin_0, end = var_1297_end_0, end_mask = var_1297_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1297_cast")];
            tensor<int32, [4]> var_1298_begin_0 = const()[name = tensor<string, []>("op_1298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1298_end_0 = const()[name = tensor<string, []>("op_1298_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1298_end_mask_0 = const()[name = tensor<string, []>("op_1298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1298_cast = slice_by_index(begin = var_1298_begin_0, end = var_1298_end_0, end_mask = var_1298_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1298_cast")];
            tensor<int32, [4]> var_1299_begin_0 = const()[name = tensor<string, []>("op_1299_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1299_end_0 = const()[name = tensor<string, []>("op_1299_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1299_end_mask_0 = const()[name = tensor<string, []>("op_1299_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1299_cast = slice_by_index(begin = var_1299_begin_0, end = var_1299_end_0, end_mask = var_1299_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1299_cast")];
            tensor<int32, [4]> var_1300_begin_0 = const()[name = tensor<string, []>("op_1300_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1300_end_0 = const()[name = tensor<string, []>("op_1300_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1300_end_mask_0 = const()[name = tensor<string, []>("op_1300_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1300_cast = slice_by_index(begin = var_1300_begin_0, end = var_1300_end_0, end_mask = var_1300_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1300_cast")];
            tensor<int32, [4]> var_1301_begin_0 = const()[name = tensor<string, []>("op_1301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1301_end_0 = const()[name = tensor<string, []>("op_1301_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1301_end_mask_0 = const()[name = tensor<string, []>("op_1301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1301_cast = slice_by_index(begin = var_1301_begin_0, end = var_1301_end_0, end_mask = var_1301_end_mask_0, x = var_1259_cast)[name = tensor<string, []>("op_1301_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_1306_begin_0 = const()[name = tensor<string, []>("op_1306_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1306_end_0 = const()[name = tensor<string, []>("op_1306_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_1306_end_mask_0 = const()[name = tensor<string, []>("op_1306_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, 64]> var_1306_cast = slice_by_index(begin = var_1306_begin_0, end = var_1306_end_0, end_mask = var_1306_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1306_cast")];
            tensor<int32, [4]> var_1310_begin_0 = const()[name = tensor<string, []>("op_1310_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1310_end_0 = const()[name = tensor<string, []>("op_1310_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_1310_end_mask_0 = const()[name = tensor<string, []>("op_1310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1310_cast = slice_by_index(begin = var_1310_begin_0, end = var_1310_end_0, end_mask = var_1310_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1310_cast")];
            tensor<int32, [4]> var_1314_begin_0 = const()[name = tensor<string, []>("op_1314_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1314_end_0 = const()[name = tensor<string, []>("op_1314_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_1314_end_mask_0 = const()[name = tensor<string, []>("op_1314_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1314_cast = slice_by_index(begin = var_1314_begin_0, end = var_1314_end_0, end_mask = var_1314_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1314_cast")];
            tensor<int32, [4]> var_1318_begin_0 = const()[name = tensor<string, []>("op_1318_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1318_end_0 = const()[name = tensor<string, []>("op_1318_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_1318_end_mask_0 = const()[name = tensor<string, []>("op_1318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1318_cast = slice_by_index(begin = var_1318_begin_0, end = var_1318_end_0, end_mask = var_1318_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1318_cast")];
            tensor<int32, [4]> var_1322_begin_0 = const()[name = tensor<string, []>("op_1322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1322_end_0 = const()[name = tensor<string, []>("op_1322_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_1322_end_mask_0 = const()[name = tensor<string, []>("op_1322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_1322_cast = slice_by_index(begin = var_1322_begin_0, end = var_1322_end_0, end_mask = var_1322_end_mask_0, x = transpose_29)[name = tensor<string, []>("op_1322_cast")];
            tensor<int32, [4]> var_1324_begin_0 = const()[name = tensor<string, []>("op_1324_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1324_end_0 = const()[name = tensor<string, []>("op_1324_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1324_end_mask_0 = const()[name = tensor<string, []>("op_1324_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1324_cast = slice_by_index(begin = var_1324_begin_0, end = var_1324_end_0, end_mask = var_1324_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1324_cast")];
            tensor<int32, [4]> var_1328_begin_0 = const()[name = tensor<string, []>("op_1328_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1328_end_0 = const()[name = tensor<string, []>("op_1328_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1328_end_mask_0 = const()[name = tensor<string, []>("op_1328_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1328_cast = slice_by_index(begin = var_1328_begin_0, end = var_1328_end_0, end_mask = var_1328_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1328_cast")];
            tensor<int32, [4]> var_1332_begin_0 = const()[name = tensor<string, []>("op_1332_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1332_end_0 = const()[name = tensor<string, []>("op_1332_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1332_end_mask_0 = const()[name = tensor<string, []>("op_1332_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1332_cast = slice_by_index(begin = var_1332_begin_0, end = var_1332_end_0, end_mask = var_1332_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1332_cast")];
            tensor<int32, [4]> var_1336_begin_0 = const()[name = tensor<string, []>("op_1336_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1336_end_0 = const()[name = tensor<string, []>("op_1336_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1336_end_mask_0 = const()[name = tensor<string, []>("op_1336_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1336_cast = slice_by_index(begin = var_1336_begin_0, end = var_1336_end_0, end_mask = var_1336_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1336_cast")];
            tensor<int32, [4]> var_1340_begin_0 = const()[name = tensor<string, []>("op_1340_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1340_end_0 = const()[name = tensor<string, []>("op_1340_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1340_end_mask_0 = const()[name = tensor<string, []>("op_1340_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1340_cast = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = v_5_cast)[name = tensor<string, []>("op_1340_cast")];
            tensor<string, []> var_1344_equation_0 = const()[name = tensor<string, []>("op_1344_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1344_cast = einsum(equation = var_1344_equation_0, values = (var_1306_cast, var_1262_cast))[name = tensor<string, []>("op_1344_cast")];
            tensor<fp16, []> var_1345_to_fp16 = const()[name = tensor<string, []>("op_1345_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_161_cast = mul(x = var_1344_cast, y = var_1345_to_fp16)[name = tensor<string, []>("aw_chunk_161_cast")];
            tensor<string, []> var_1348_equation_0 = const()[name = tensor<string, []>("op_1348_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1348_cast = einsum(equation = var_1348_equation_0, values = (var_1306_cast, var_1263_cast))[name = tensor<string, []>("op_1348_cast")];
            tensor<fp16, []> var_1349_to_fp16 = const()[name = tensor<string, []>("op_1349_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_163_cast = mul(x = var_1348_cast, y = var_1349_to_fp16)[name = tensor<string, []>("aw_chunk_163_cast")];
            tensor<string, []> var_1352_equation_0 = const()[name = tensor<string, []>("op_1352_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1352_cast = einsum(equation = var_1352_equation_0, values = (var_1306_cast, var_1264_cast))[name = tensor<string, []>("op_1352_cast")];
            tensor<fp16, []> var_1353_to_fp16 = const()[name = tensor<string, []>("op_1353_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_165_cast = mul(x = var_1352_cast, y = var_1353_to_fp16)[name = tensor<string, []>("aw_chunk_165_cast")];
            tensor<string, []> var_1356_equation_0 = const()[name = tensor<string, []>("op_1356_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1356_cast = einsum(equation = var_1356_equation_0, values = (var_1306_cast, var_1265_cast))[name = tensor<string, []>("op_1356_cast")];
            tensor<fp16, []> var_1357_to_fp16 = const()[name = tensor<string, []>("op_1357_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_167_cast = mul(x = var_1356_cast, y = var_1357_to_fp16)[name = tensor<string, []>("aw_chunk_167_cast")];
            tensor<string, []> var_1360_equation_0 = const()[name = tensor<string, []>("op_1360_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1360_cast = einsum(equation = var_1360_equation_0, values = (var_1306_cast, var_1266_cast))[name = tensor<string, []>("op_1360_cast")];
            tensor<fp16, []> var_1361_to_fp16 = const()[name = tensor<string, []>("op_1361_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_169_cast = mul(x = var_1360_cast, y = var_1361_to_fp16)[name = tensor<string, []>("aw_chunk_169_cast")];
            tensor<string, []> var_1364_equation_0 = const()[name = tensor<string, []>("op_1364_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1364_cast = einsum(equation = var_1364_equation_0, values = (var_1306_cast, var_1267_cast))[name = tensor<string, []>("op_1364_cast")];
            tensor<fp16, []> var_1365_to_fp16 = const()[name = tensor<string, []>("op_1365_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_171_cast = mul(x = var_1364_cast, y = var_1365_to_fp16)[name = tensor<string, []>("aw_chunk_171_cast")];
            tensor<string, []> var_1368_equation_0 = const()[name = tensor<string, []>("op_1368_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1368_cast = einsum(equation = var_1368_equation_0, values = (var_1306_cast, var_1268_cast))[name = tensor<string, []>("op_1368_cast")];
            tensor<fp16, []> var_1369_to_fp16 = const()[name = tensor<string, []>("op_1369_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_173_cast = mul(x = var_1368_cast, y = var_1369_to_fp16)[name = tensor<string, []>("aw_chunk_173_cast")];
            tensor<string, []> var_1372_equation_0 = const()[name = tensor<string, []>("op_1372_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1372_cast = einsum(equation = var_1372_equation_0, values = (var_1306_cast, var_1269_cast))[name = tensor<string, []>("op_1372_cast")];
            tensor<fp16, []> var_1373_to_fp16 = const()[name = tensor<string, []>("op_1373_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_175_cast = mul(x = var_1372_cast, y = var_1373_to_fp16)[name = tensor<string, []>("aw_chunk_175_cast")];
            tensor<string, []> var_1376_equation_0 = const()[name = tensor<string, []>("op_1376_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1376_cast = einsum(equation = var_1376_equation_0, values = (var_1310_cast, var_1270_cast))[name = tensor<string, []>("op_1376_cast")];
            tensor<fp16, []> var_1377_to_fp16 = const()[name = tensor<string, []>("op_1377_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_177_cast = mul(x = var_1376_cast, y = var_1377_to_fp16)[name = tensor<string, []>("aw_chunk_177_cast")];
            tensor<string, []> var_1380_equation_0 = const()[name = tensor<string, []>("op_1380_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1380_cast = einsum(equation = var_1380_equation_0, values = (var_1310_cast, var_1271_cast))[name = tensor<string, []>("op_1380_cast")];
            tensor<fp16, []> var_1381_to_fp16 = const()[name = tensor<string, []>("op_1381_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_179_cast = mul(x = var_1380_cast, y = var_1381_to_fp16)[name = tensor<string, []>("aw_chunk_179_cast")];
            tensor<string, []> var_1384_equation_0 = const()[name = tensor<string, []>("op_1384_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1384_cast = einsum(equation = var_1384_equation_0, values = (var_1310_cast, var_1272_cast))[name = tensor<string, []>("op_1384_cast")];
            tensor<fp16, []> var_1385_to_fp16 = const()[name = tensor<string, []>("op_1385_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_181_cast = mul(x = var_1384_cast, y = var_1385_to_fp16)[name = tensor<string, []>("aw_chunk_181_cast")];
            tensor<string, []> var_1388_equation_0 = const()[name = tensor<string, []>("op_1388_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1388_cast = einsum(equation = var_1388_equation_0, values = (var_1310_cast, var_1273_cast))[name = tensor<string, []>("op_1388_cast")];
            tensor<fp16, []> var_1389_to_fp16 = const()[name = tensor<string, []>("op_1389_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_183_cast = mul(x = var_1388_cast, y = var_1389_to_fp16)[name = tensor<string, []>("aw_chunk_183_cast")];
            tensor<string, []> var_1392_equation_0 = const()[name = tensor<string, []>("op_1392_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1392_cast = einsum(equation = var_1392_equation_0, values = (var_1310_cast, var_1274_cast))[name = tensor<string, []>("op_1392_cast")];
            tensor<fp16, []> var_1393_to_fp16 = const()[name = tensor<string, []>("op_1393_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_185_cast = mul(x = var_1392_cast, y = var_1393_to_fp16)[name = tensor<string, []>("aw_chunk_185_cast")];
            tensor<string, []> var_1396_equation_0 = const()[name = tensor<string, []>("op_1396_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1396_cast = einsum(equation = var_1396_equation_0, values = (var_1310_cast, var_1275_cast))[name = tensor<string, []>("op_1396_cast")];
            tensor<fp16, []> var_1397_to_fp16 = const()[name = tensor<string, []>("op_1397_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_187_cast = mul(x = var_1396_cast, y = var_1397_to_fp16)[name = tensor<string, []>("aw_chunk_187_cast")];
            tensor<string, []> var_1400_equation_0 = const()[name = tensor<string, []>("op_1400_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1400_cast = einsum(equation = var_1400_equation_0, values = (var_1310_cast, var_1276_cast))[name = tensor<string, []>("op_1400_cast")];
            tensor<fp16, []> var_1401_to_fp16 = const()[name = tensor<string, []>("op_1401_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_189_cast = mul(x = var_1400_cast, y = var_1401_to_fp16)[name = tensor<string, []>("aw_chunk_189_cast")];
            tensor<string, []> var_1404_equation_0 = const()[name = tensor<string, []>("op_1404_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1404_cast = einsum(equation = var_1404_equation_0, values = (var_1310_cast, var_1277_cast))[name = tensor<string, []>("op_1404_cast")];
            tensor<fp16, []> var_1405_to_fp16 = const()[name = tensor<string, []>("op_1405_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_191_cast = mul(x = var_1404_cast, y = var_1405_to_fp16)[name = tensor<string, []>("aw_chunk_191_cast")];
            tensor<string, []> var_1408_equation_0 = const()[name = tensor<string, []>("op_1408_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1408_cast = einsum(equation = var_1408_equation_0, values = (var_1314_cast, var_1278_cast))[name = tensor<string, []>("op_1408_cast")];
            tensor<fp16, []> var_1409_to_fp16 = const()[name = tensor<string, []>("op_1409_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_193_cast = mul(x = var_1408_cast, y = var_1409_to_fp16)[name = tensor<string, []>("aw_chunk_193_cast")];
            tensor<string, []> var_1412_equation_0 = const()[name = tensor<string, []>("op_1412_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1412_cast = einsum(equation = var_1412_equation_0, values = (var_1314_cast, var_1279_cast))[name = tensor<string, []>("op_1412_cast")];
            tensor<fp16, []> var_1413_to_fp16 = const()[name = tensor<string, []>("op_1413_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_195_cast = mul(x = var_1412_cast, y = var_1413_to_fp16)[name = tensor<string, []>("aw_chunk_195_cast")];
            tensor<string, []> var_1416_equation_0 = const()[name = tensor<string, []>("op_1416_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1416_cast = einsum(equation = var_1416_equation_0, values = (var_1314_cast, var_1280_cast))[name = tensor<string, []>("op_1416_cast")];
            tensor<fp16, []> var_1417_to_fp16 = const()[name = tensor<string, []>("op_1417_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_197_cast = mul(x = var_1416_cast, y = var_1417_to_fp16)[name = tensor<string, []>("aw_chunk_197_cast")];
            tensor<string, []> var_1420_equation_0 = const()[name = tensor<string, []>("op_1420_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1420_cast = einsum(equation = var_1420_equation_0, values = (var_1314_cast, var_1281_cast))[name = tensor<string, []>("op_1420_cast")];
            tensor<fp16, []> var_1421_to_fp16 = const()[name = tensor<string, []>("op_1421_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_199_cast = mul(x = var_1420_cast, y = var_1421_to_fp16)[name = tensor<string, []>("aw_chunk_199_cast")];
            tensor<string, []> var_1424_equation_0 = const()[name = tensor<string, []>("op_1424_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1424_cast = einsum(equation = var_1424_equation_0, values = (var_1314_cast, var_1282_cast))[name = tensor<string, []>("op_1424_cast")];
            tensor<fp16, []> var_1425_to_fp16 = const()[name = tensor<string, []>("op_1425_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_201_cast = mul(x = var_1424_cast, y = var_1425_to_fp16)[name = tensor<string, []>("aw_chunk_201_cast")];
            tensor<string, []> var_1428_equation_0 = const()[name = tensor<string, []>("op_1428_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1428_cast = einsum(equation = var_1428_equation_0, values = (var_1314_cast, var_1283_cast))[name = tensor<string, []>("op_1428_cast")];
            tensor<fp16, []> var_1429_to_fp16 = const()[name = tensor<string, []>("op_1429_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_203_cast = mul(x = var_1428_cast, y = var_1429_to_fp16)[name = tensor<string, []>("aw_chunk_203_cast")];
            tensor<string, []> var_1432_equation_0 = const()[name = tensor<string, []>("op_1432_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1432_cast = einsum(equation = var_1432_equation_0, values = (var_1314_cast, var_1284_cast))[name = tensor<string, []>("op_1432_cast")];
            tensor<fp16, []> var_1433_to_fp16 = const()[name = tensor<string, []>("op_1433_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_205_cast = mul(x = var_1432_cast, y = var_1433_to_fp16)[name = tensor<string, []>("aw_chunk_205_cast")];
            tensor<string, []> var_1436_equation_0 = const()[name = tensor<string, []>("op_1436_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1436_cast = einsum(equation = var_1436_equation_0, values = (var_1314_cast, var_1285_cast))[name = tensor<string, []>("op_1436_cast")];
            tensor<fp16, []> var_1437_to_fp16 = const()[name = tensor<string, []>("op_1437_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_207_cast = mul(x = var_1436_cast, y = var_1437_to_fp16)[name = tensor<string, []>("aw_chunk_207_cast")];
            tensor<string, []> var_1440_equation_0 = const()[name = tensor<string, []>("op_1440_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1440_cast = einsum(equation = var_1440_equation_0, values = (var_1318_cast, var_1286_cast))[name = tensor<string, []>("op_1440_cast")];
            tensor<fp16, []> var_1441_to_fp16 = const()[name = tensor<string, []>("op_1441_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_209_cast = mul(x = var_1440_cast, y = var_1441_to_fp16)[name = tensor<string, []>("aw_chunk_209_cast")];
            tensor<string, []> var_1444_equation_0 = const()[name = tensor<string, []>("op_1444_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1444_cast = einsum(equation = var_1444_equation_0, values = (var_1318_cast, var_1287_cast))[name = tensor<string, []>("op_1444_cast")];
            tensor<fp16, []> var_1445_to_fp16 = const()[name = tensor<string, []>("op_1445_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_211_cast = mul(x = var_1444_cast, y = var_1445_to_fp16)[name = tensor<string, []>("aw_chunk_211_cast")];
            tensor<string, []> var_1448_equation_0 = const()[name = tensor<string, []>("op_1448_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1448_cast = einsum(equation = var_1448_equation_0, values = (var_1318_cast, var_1288_cast))[name = tensor<string, []>("op_1448_cast")];
            tensor<fp16, []> var_1449_to_fp16 = const()[name = tensor<string, []>("op_1449_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_213_cast = mul(x = var_1448_cast, y = var_1449_to_fp16)[name = tensor<string, []>("aw_chunk_213_cast")];
            tensor<string, []> var_1452_equation_0 = const()[name = tensor<string, []>("op_1452_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1452_cast = einsum(equation = var_1452_equation_0, values = (var_1318_cast, var_1289_cast))[name = tensor<string, []>("op_1452_cast")];
            tensor<fp16, []> var_1453_to_fp16 = const()[name = tensor<string, []>("op_1453_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_215_cast = mul(x = var_1452_cast, y = var_1453_to_fp16)[name = tensor<string, []>("aw_chunk_215_cast")];
            tensor<string, []> var_1456_equation_0 = const()[name = tensor<string, []>("op_1456_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1456_cast = einsum(equation = var_1456_equation_0, values = (var_1318_cast, var_1290_cast))[name = tensor<string, []>("op_1456_cast")];
            tensor<fp16, []> var_1457_to_fp16 = const()[name = tensor<string, []>("op_1457_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_217_cast = mul(x = var_1456_cast, y = var_1457_to_fp16)[name = tensor<string, []>("aw_chunk_217_cast")];
            tensor<string, []> var_1460_equation_0 = const()[name = tensor<string, []>("op_1460_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1460_cast = einsum(equation = var_1460_equation_0, values = (var_1318_cast, var_1291_cast))[name = tensor<string, []>("op_1460_cast")];
            tensor<fp16, []> var_1461_to_fp16 = const()[name = tensor<string, []>("op_1461_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_219_cast = mul(x = var_1460_cast, y = var_1461_to_fp16)[name = tensor<string, []>("aw_chunk_219_cast")];
            tensor<string, []> var_1464_equation_0 = const()[name = tensor<string, []>("op_1464_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1464_cast = einsum(equation = var_1464_equation_0, values = (var_1318_cast, var_1292_cast))[name = tensor<string, []>("op_1464_cast")];
            tensor<fp16, []> var_1465_to_fp16 = const()[name = tensor<string, []>("op_1465_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_221_cast = mul(x = var_1464_cast, y = var_1465_to_fp16)[name = tensor<string, []>("aw_chunk_221_cast")];
            tensor<string, []> var_1468_equation_0 = const()[name = tensor<string, []>("op_1468_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1468_cast = einsum(equation = var_1468_equation_0, values = (var_1318_cast, var_1293_cast))[name = tensor<string, []>("op_1468_cast")];
            tensor<fp16, []> var_1469_to_fp16 = const()[name = tensor<string, []>("op_1469_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_223_cast = mul(x = var_1468_cast, y = var_1469_to_fp16)[name = tensor<string, []>("aw_chunk_223_cast")];
            tensor<string, []> var_1472_equation_0 = const()[name = tensor<string, []>("op_1472_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1472_cast = einsum(equation = var_1472_equation_0, values = (var_1322_cast, var_1294_cast))[name = tensor<string, []>("op_1472_cast")];
            tensor<fp16, []> var_1473_to_fp16 = const()[name = tensor<string, []>("op_1473_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_225_cast = mul(x = var_1472_cast, y = var_1473_to_fp16)[name = tensor<string, []>("aw_chunk_225_cast")];
            tensor<string, []> var_1476_equation_0 = const()[name = tensor<string, []>("op_1476_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1476_cast = einsum(equation = var_1476_equation_0, values = (var_1322_cast, var_1295_cast))[name = tensor<string, []>("op_1476_cast")];
            tensor<fp16, []> var_1477_to_fp16 = const()[name = tensor<string, []>("op_1477_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_227_cast = mul(x = var_1476_cast, y = var_1477_to_fp16)[name = tensor<string, []>("aw_chunk_227_cast")];
            tensor<string, []> var_1480_equation_0 = const()[name = tensor<string, []>("op_1480_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1480_cast = einsum(equation = var_1480_equation_0, values = (var_1322_cast, var_1296_cast))[name = tensor<string, []>("op_1480_cast")];
            tensor<fp16, []> var_1481_to_fp16 = const()[name = tensor<string, []>("op_1481_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_229_cast = mul(x = var_1480_cast, y = var_1481_to_fp16)[name = tensor<string, []>("aw_chunk_229_cast")];
            tensor<string, []> var_1484_equation_0 = const()[name = tensor<string, []>("op_1484_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1484_cast = einsum(equation = var_1484_equation_0, values = (var_1322_cast, var_1297_cast))[name = tensor<string, []>("op_1484_cast")];
            tensor<fp16, []> var_1485_to_fp16 = const()[name = tensor<string, []>("op_1485_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_231_cast = mul(x = var_1484_cast, y = var_1485_to_fp16)[name = tensor<string, []>("aw_chunk_231_cast")];
            tensor<string, []> var_1488_equation_0 = const()[name = tensor<string, []>("op_1488_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1488_cast = einsum(equation = var_1488_equation_0, values = (var_1322_cast, var_1298_cast))[name = tensor<string, []>("op_1488_cast")];
            tensor<fp16, []> var_1489_to_fp16 = const()[name = tensor<string, []>("op_1489_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_233_cast = mul(x = var_1488_cast, y = var_1489_to_fp16)[name = tensor<string, []>("aw_chunk_233_cast")];
            tensor<string, []> var_1492_equation_0 = const()[name = tensor<string, []>("op_1492_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1492_cast = einsum(equation = var_1492_equation_0, values = (var_1322_cast, var_1299_cast))[name = tensor<string, []>("op_1492_cast")];
            tensor<fp16, []> var_1493_to_fp16 = const()[name = tensor<string, []>("op_1493_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_235_cast = mul(x = var_1492_cast, y = var_1493_to_fp16)[name = tensor<string, []>("aw_chunk_235_cast")];
            tensor<string, []> var_1496_equation_0 = const()[name = tensor<string, []>("op_1496_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1496_cast = einsum(equation = var_1496_equation_0, values = (var_1322_cast, var_1300_cast))[name = tensor<string, []>("op_1496_cast")];
            tensor<fp16, []> var_1497_to_fp16 = const()[name = tensor<string, []>("op_1497_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_237_cast = mul(x = var_1496_cast, y = var_1497_to_fp16)[name = tensor<string, []>("aw_chunk_237_cast")];
            tensor<string, []> var_1500_equation_0 = const()[name = tensor<string, []>("op_1500_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_1500_cast = einsum(equation = var_1500_equation_0, values = (var_1322_cast, var_1301_cast))[name = tensor<string, []>("op_1500_cast")];
            tensor<fp16, []> var_1501_to_fp16 = const()[name = tensor<string, []>("op_1501_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_239_cast = mul(x = var_1500_cast, y = var_1501_to_fp16)[name = tensor<string, []>("aw_chunk_239_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1503_cast = softmax(axis = var_120, x = aw_chunk_161_cast)[name = tensor<string, []>("op_1503_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1504_cast = softmax(axis = var_120, x = aw_chunk_163_cast)[name = tensor<string, []>("op_1504_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1505_cast = softmax(axis = var_120, x = aw_chunk_165_cast)[name = tensor<string, []>("op_1505_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1506_cast = softmax(axis = var_120, x = aw_chunk_167_cast)[name = tensor<string, []>("op_1506_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1507_cast = softmax(axis = var_120, x = aw_chunk_169_cast)[name = tensor<string, []>("op_1507_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1508_cast = softmax(axis = var_120, x = aw_chunk_171_cast)[name = tensor<string, []>("op_1508_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1509_cast = softmax(axis = var_120, x = aw_chunk_173_cast)[name = tensor<string, []>("op_1509_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1510_cast = softmax(axis = var_120, x = aw_chunk_175_cast)[name = tensor<string, []>("op_1510_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1511_cast = softmax(axis = var_120, x = aw_chunk_177_cast)[name = tensor<string, []>("op_1511_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1512_cast = softmax(axis = var_120, x = aw_chunk_179_cast)[name = tensor<string, []>("op_1512_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1513_cast = softmax(axis = var_120, x = aw_chunk_181_cast)[name = tensor<string, []>("op_1513_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1514_cast = softmax(axis = var_120, x = aw_chunk_183_cast)[name = tensor<string, []>("op_1514_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1515_cast = softmax(axis = var_120, x = aw_chunk_185_cast)[name = tensor<string, []>("op_1515_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1516_cast = softmax(axis = var_120, x = aw_chunk_187_cast)[name = tensor<string, []>("op_1516_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1517_cast = softmax(axis = var_120, x = aw_chunk_189_cast)[name = tensor<string, []>("op_1517_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1518_cast = softmax(axis = var_120, x = aw_chunk_191_cast)[name = tensor<string, []>("op_1518_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1519_cast = softmax(axis = var_120, x = aw_chunk_193_cast)[name = tensor<string, []>("op_1519_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1520_cast = softmax(axis = var_120, x = aw_chunk_195_cast)[name = tensor<string, []>("op_1520_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1521_cast = softmax(axis = var_120, x = aw_chunk_197_cast)[name = tensor<string, []>("op_1521_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1522_cast = softmax(axis = var_120, x = aw_chunk_199_cast)[name = tensor<string, []>("op_1522_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1523_cast = softmax(axis = var_120, x = aw_chunk_201_cast)[name = tensor<string, []>("op_1523_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1524_cast = softmax(axis = var_120, x = aw_chunk_203_cast)[name = tensor<string, []>("op_1524_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1525_cast = softmax(axis = var_120, x = aw_chunk_205_cast)[name = tensor<string, []>("op_1525_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1526_cast = softmax(axis = var_120, x = aw_chunk_207_cast)[name = tensor<string, []>("op_1526_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1527_cast = softmax(axis = var_120, x = aw_chunk_209_cast)[name = tensor<string, []>("op_1527_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1528_cast = softmax(axis = var_120, x = aw_chunk_211_cast)[name = tensor<string, []>("op_1528_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1529_cast = softmax(axis = var_120, x = aw_chunk_213_cast)[name = tensor<string, []>("op_1529_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1530_cast = softmax(axis = var_120, x = aw_chunk_215_cast)[name = tensor<string, []>("op_1530_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1531_cast = softmax(axis = var_120, x = aw_chunk_217_cast)[name = tensor<string, []>("op_1531_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1532_cast = softmax(axis = var_120, x = aw_chunk_219_cast)[name = tensor<string, []>("op_1532_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1533_cast = softmax(axis = var_120, x = aw_chunk_221_cast)[name = tensor<string, []>("op_1533_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1534_cast = softmax(axis = var_120, x = aw_chunk_223_cast)[name = tensor<string, []>("op_1534_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1535_cast = softmax(axis = var_120, x = aw_chunk_225_cast)[name = tensor<string, []>("op_1535_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1536_cast = softmax(axis = var_120, x = aw_chunk_227_cast)[name = tensor<string, []>("op_1536_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1537_cast = softmax(axis = var_120, x = aw_chunk_229_cast)[name = tensor<string, []>("op_1537_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1538_cast = softmax(axis = var_120, x = aw_chunk_231_cast)[name = tensor<string, []>("op_1538_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1539_cast = softmax(axis = var_120, x = aw_chunk_233_cast)[name = tensor<string, []>("op_1539_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1540_cast = softmax(axis = var_120, x = aw_chunk_235_cast)[name = tensor<string, []>("op_1540_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1541_cast = softmax(axis = var_120, x = aw_chunk_237_cast)[name = tensor<string, []>("op_1541_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_1542_cast = softmax(axis = var_120, x = aw_chunk_239_cast)[name = tensor<string, []>("op_1542_cast")];
            tensor<string, []> var_1544_equation_0 = const()[name = tensor<string, []>("op_1544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1544_cast = einsum(equation = var_1544_equation_0, values = (var_1324_cast, var_1503_cast))[name = tensor<string, []>("op_1544_cast")];
            tensor<string, []> var_1546_equation_0 = const()[name = tensor<string, []>("op_1546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1546_cast = einsum(equation = var_1546_equation_0, values = (var_1324_cast, var_1504_cast))[name = tensor<string, []>("op_1546_cast")];
            tensor<string, []> var_1548_equation_0 = const()[name = tensor<string, []>("op_1548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1548_cast = einsum(equation = var_1548_equation_0, values = (var_1324_cast, var_1505_cast))[name = tensor<string, []>("op_1548_cast")];
            tensor<string, []> var_1550_equation_0 = const()[name = tensor<string, []>("op_1550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1550_cast = einsum(equation = var_1550_equation_0, values = (var_1324_cast, var_1506_cast))[name = tensor<string, []>("op_1550_cast")];
            tensor<string, []> var_1552_equation_0 = const()[name = tensor<string, []>("op_1552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1552_cast = einsum(equation = var_1552_equation_0, values = (var_1324_cast, var_1507_cast))[name = tensor<string, []>("op_1552_cast")];
            tensor<string, []> var_1554_equation_0 = const()[name = tensor<string, []>("op_1554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1554_cast = einsum(equation = var_1554_equation_0, values = (var_1324_cast, var_1508_cast))[name = tensor<string, []>("op_1554_cast")];
            tensor<string, []> var_1556_equation_0 = const()[name = tensor<string, []>("op_1556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1556_cast = einsum(equation = var_1556_equation_0, values = (var_1324_cast, var_1509_cast))[name = tensor<string, []>("op_1556_cast")];
            tensor<string, []> var_1558_equation_0 = const()[name = tensor<string, []>("op_1558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1558_cast = einsum(equation = var_1558_equation_0, values = (var_1324_cast, var_1510_cast))[name = tensor<string, []>("op_1558_cast")];
            tensor<string, []> var_1560_equation_0 = const()[name = tensor<string, []>("op_1560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1560_cast = einsum(equation = var_1560_equation_0, values = (var_1328_cast, var_1511_cast))[name = tensor<string, []>("op_1560_cast")];
            tensor<string, []> var_1562_equation_0 = const()[name = tensor<string, []>("op_1562_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1562_cast = einsum(equation = var_1562_equation_0, values = (var_1328_cast, var_1512_cast))[name = tensor<string, []>("op_1562_cast")];
            tensor<string, []> var_1564_equation_0 = const()[name = tensor<string, []>("op_1564_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1564_cast = einsum(equation = var_1564_equation_0, values = (var_1328_cast, var_1513_cast))[name = tensor<string, []>("op_1564_cast")];
            tensor<string, []> var_1566_equation_0 = const()[name = tensor<string, []>("op_1566_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1566_cast = einsum(equation = var_1566_equation_0, values = (var_1328_cast, var_1514_cast))[name = tensor<string, []>("op_1566_cast")];
            tensor<string, []> var_1568_equation_0 = const()[name = tensor<string, []>("op_1568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1568_cast = einsum(equation = var_1568_equation_0, values = (var_1328_cast, var_1515_cast))[name = tensor<string, []>("op_1568_cast")];
            tensor<string, []> var_1570_equation_0 = const()[name = tensor<string, []>("op_1570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1570_cast = einsum(equation = var_1570_equation_0, values = (var_1328_cast, var_1516_cast))[name = tensor<string, []>("op_1570_cast")];
            tensor<string, []> var_1572_equation_0 = const()[name = tensor<string, []>("op_1572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1572_cast = einsum(equation = var_1572_equation_0, values = (var_1328_cast, var_1517_cast))[name = tensor<string, []>("op_1572_cast")];
            tensor<string, []> var_1574_equation_0 = const()[name = tensor<string, []>("op_1574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1574_cast = einsum(equation = var_1574_equation_0, values = (var_1328_cast, var_1518_cast))[name = tensor<string, []>("op_1574_cast")];
            tensor<string, []> var_1576_equation_0 = const()[name = tensor<string, []>("op_1576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1576_cast = einsum(equation = var_1576_equation_0, values = (var_1332_cast, var_1519_cast))[name = tensor<string, []>("op_1576_cast")];
            tensor<string, []> var_1578_equation_0 = const()[name = tensor<string, []>("op_1578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1578_cast = einsum(equation = var_1578_equation_0, values = (var_1332_cast, var_1520_cast))[name = tensor<string, []>("op_1578_cast")];
            tensor<string, []> var_1580_equation_0 = const()[name = tensor<string, []>("op_1580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1580_cast = einsum(equation = var_1580_equation_0, values = (var_1332_cast, var_1521_cast))[name = tensor<string, []>("op_1580_cast")];
            tensor<string, []> var_1582_equation_0 = const()[name = tensor<string, []>("op_1582_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1582_cast = einsum(equation = var_1582_equation_0, values = (var_1332_cast, var_1522_cast))[name = tensor<string, []>("op_1582_cast")];
            tensor<string, []> var_1584_equation_0 = const()[name = tensor<string, []>("op_1584_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1584_cast = einsum(equation = var_1584_equation_0, values = (var_1332_cast, var_1523_cast))[name = tensor<string, []>("op_1584_cast")];
            tensor<string, []> var_1586_equation_0 = const()[name = tensor<string, []>("op_1586_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1586_cast = einsum(equation = var_1586_equation_0, values = (var_1332_cast, var_1524_cast))[name = tensor<string, []>("op_1586_cast")];
            tensor<string, []> var_1588_equation_0 = const()[name = tensor<string, []>("op_1588_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1588_cast = einsum(equation = var_1588_equation_0, values = (var_1332_cast, var_1525_cast))[name = tensor<string, []>("op_1588_cast")];
            tensor<string, []> var_1590_equation_0 = const()[name = tensor<string, []>("op_1590_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1590_cast = einsum(equation = var_1590_equation_0, values = (var_1332_cast, var_1526_cast))[name = tensor<string, []>("op_1590_cast")];
            tensor<string, []> var_1592_equation_0 = const()[name = tensor<string, []>("op_1592_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1592_cast = einsum(equation = var_1592_equation_0, values = (var_1336_cast, var_1527_cast))[name = tensor<string, []>("op_1592_cast")];
            tensor<string, []> var_1594_equation_0 = const()[name = tensor<string, []>("op_1594_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1594_cast = einsum(equation = var_1594_equation_0, values = (var_1336_cast, var_1528_cast))[name = tensor<string, []>("op_1594_cast")];
            tensor<string, []> var_1596_equation_0 = const()[name = tensor<string, []>("op_1596_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1596_cast = einsum(equation = var_1596_equation_0, values = (var_1336_cast, var_1529_cast))[name = tensor<string, []>("op_1596_cast")];
            tensor<string, []> var_1598_equation_0 = const()[name = tensor<string, []>("op_1598_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1598_cast = einsum(equation = var_1598_equation_0, values = (var_1336_cast, var_1530_cast))[name = tensor<string, []>("op_1598_cast")];
            tensor<string, []> var_1600_equation_0 = const()[name = tensor<string, []>("op_1600_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1600_cast = einsum(equation = var_1600_equation_0, values = (var_1336_cast, var_1531_cast))[name = tensor<string, []>("op_1600_cast")];
            tensor<string, []> var_1602_equation_0 = const()[name = tensor<string, []>("op_1602_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1602_cast = einsum(equation = var_1602_equation_0, values = (var_1336_cast, var_1532_cast))[name = tensor<string, []>("op_1602_cast")];
            tensor<string, []> var_1604_equation_0 = const()[name = tensor<string, []>("op_1604_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1604_cast = einsum(equation = var_1604_equation_0, values = (var_1336_cast, var_1533_cast))[name = tensor<string, []>("op_1604_cast")];
            tensor<string, []> var_1606_equation_0 = const()[name = tensor<string, []>("op_1606_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1606_cast = einsum(equation = var_1606_equation_0, values = (var_1336_cast, var_1534_cast))[name = tensor<string, []>("op_1606_cast")];
            tensor<string, []> var_1608_equation_0 = const()[name = tensor<string, []>("op_1608_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1608_cast = einsum(equation = var_1608_equation_0, values = (var_1340_cast, var_1535_cast))[name = tensor<string, []>("op_1608_cast")];
            tensor<string, []> var_1610_equation_0 = const()[name = tensor<string, []>("op_1610_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1610_cast = einsum(equation = var_1610_equation_0, values = (var_1340_cast, var_1536_cast))[name = tensor<string, []>("op_1610_cast")];
            tensor<string, []> var_1612_equation_0 = const()[name = tensor<string, []>("op_1612_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1612_cast = einsum(equation = var_1612_equation_0, values = (var_1340_cast, var_1537_cast))[name = tensor<string, []>("op_1612_cast")];
            tensor<string, []> var_1614_equation_0 = const()[name = tensor<string, []>("op_1614_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1614_cast = einsum(equation = var_1614_equation_0, values = (var_1340_cast, var_1538_cast))[name = tensor<string, []>("op_1614_cast")];
            tensor<string, []> var_1616_equation_0 = const()[name = tensor<string, []>("op_1616_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1616_cast = einsum(equation = var_1616_equation_0, values = (var_1340_cast, var_1539_cast))[name = tensor<string, []>("op_1616_cast")];
            tensor<string, []> var_1618_equation_0 = const()[name = tensor<string, []>("op_1618_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1618_cast = einsum(equation = var_1618_equation_0, values = (var_1340_cast, var_1540_cast))[name = tensor<string, []>("op_1618_cast")];
            tensor<string, []> var_1620_equation_0 = const()[name = tensor<string, []>("op_1620_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1620_cast = einsum(equation = var_1620_equation_0, values = (var_1340_cast, var_1541_cast))[name = tensor<string, []>("op_1620_cast")];
            tensor<string, []> var_1622_equation_0 = const()[name = tensor<string, []>("op_1622_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1622_cast = einsum(equation = var_1622_equation_0, values = (var_1340_cast, var_1542_cast))[name = tensor<string, []>("op_1622_cast")];
            tensor<bool, []> var_1624_interleave_0 = const()[name = tensor<string, []>("op_1624_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1624_cast = concat(axis = var_95, interleave = var_1624_interleave_0, values = (var_1544_cast, var_1546_cast, var_1548_cast, var_1550_cast, var_1552_cast, var_1554_cast, var_1556_cast, var_1558_cast))[name = tensor<string, []>("op_1624_cast")];
            tensor<bool, []> var_1626_interleave_0 = const()[name = tensor<string, []>("op_1626_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1626_cast = concat(axis = var_95, interleave = var_1626_interleave_0, values = (var_1560_cast, var_1562_cast, var_1564_cast, var_1566_cast, var_1568_cast, var_1570_cast, var_1572_cast, var_1574_cast))[name = tensor<string, []>("op_1626_cast")];
            tensor<bool, []> var_1628_interleave_0 = const()[name = tensor<string, []>("op_1628_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1628_cast = concat(axis = var_95, interleave = var_1628_interleave_0, values = (var_1576_cast, var_1578_cast, var_1580_cast, var_1582_cast, var_1584_cast, var_1586_cast, var_1588_cast, var_1590_cast))[name = tensor<string, []>("op_1628_cast")];
            tensor<bool, []> var_1630_interleave_0 = const()[name = tensor<string, []>("op_1630_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1630_cast = concat(axis = var_95, interleave = var_1630_interleave_0, values = (var_1592_cast, var_1594_cast, var_1596_cast, var_1598_cast, var_1600_cast, var_1602_cast, var_1604_cast, var_1606_cast))[name = tensor<string, []>("op_1630_cast")];
            tensor<bool, []> var_1632_interleave_0 = const()[name = tensor<string, []>("op_1632_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_1632_cast = concat(axis = var_95, interleave = var_1632_interleave_0, values = (var_1608_cast, var_1610_cast, var_1612_cast, var_1614_cast, var_1616_cast, var_1618_cast, var_1620_cast, var_1622_cast))[name = tensor<string, []>("op_1632_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_120, interleave = input_51_interleave_0, values = (var_1624_cast, var_1626_cast, var_1628_cast, var_1630_cast, var_1632_cast))[name = tensor<string, []>("input_51_cast")];
            tensor<int32, [2]> var_1638 = const()[name = tensor<string, []>("op_1638"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1640 = const()[name = tensor<string, []>("op_1640"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_1642_pad_type_0 = const()[name = tensor<string, []>("op_1642_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_1642_pad_0 = const()[name = tensor<string, []>("op_1642_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2452160))), lut = tensor<fp16, [4]>([-0x1.3f4p-4, -0x1.794p-6, 0x1.734p-6, 0x1.3ccp-4]), 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, []>(2477824)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1642_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_1640, groups = var_120, pad = var_1642_pad_0, pad_type = var_1642_pad_type_0, strides = var_1638, 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_1642_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_9_cast = add(x = var_1642_cast, y = inputs_7_cast)[name = tensor<string, []>("inputs_9_cast")];
            tensor<int32, [1]> var_1646 = const()[name = tensor<string, []>("op_1646"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_9_cast = reduce_mean(axes = var_1646, keep_dims = var_115, 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_1650 = const()[name = tensor<string, []>("op_1650"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_1651_cast = reduce_mean(axes = var_1650, keep_dims = var_115, x = zero_mean_sq_9_cast)[name = tensor<string, []>("op_1651_cast")];
            tensor<fp16, []> var_1652_to_fp16 = const()[name = tensor<string, []>("op_1652_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_1653_cast = add(x = var_1651_cast, y = var_1652_to_fp16)[name = tensor<string, []>("op_1653_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_1653_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_1657_to_fp16 = const()[name = tensor<string, []>("op_1657_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2478528)))];
            tensor<fp16, [2, 320, 1, 4096]> var_1658_cast = add(x = out_9_cast, y = var_1657_to_fp16)[name = tensor<string, []>("op_1658_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_1660_to_fp16 = const()[name = tensor<string, []>("op_1660_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2479232)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_29_cast = mul(x = var_1658_cast, y = var_1660_to_fp16)[name = tensor<string, []>("hidden_states_29_cast")];
            tensor<int32, [2]> var_1667 = const()[name = tensor<string, []>("op_1667"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1669 = const()[name = tensor<string, []>("op_1669"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2479936))), lut = tensor<fp16, [4]>([-0x1.854p-4, -0x1.d14p-6, 0x1.ccp-6, 0x1.818p-4]), 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_1669, groups = var_120, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = var_1667, 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_1673 = const()[name = tensor<string, []>("op_1673"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1675 = const()[name = tensor<string, []>("op_1675"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_13_pad_type_0 = const()[name = tensor<string, []>("k_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_13_pad_0 = const()[name = tensor<string, []>("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2505600))), lut = tensor<fp16, [4]>([-0x1.33p-4, -0x1.5f8p-6, 0x1.694p-6, 0x1.368p-4]), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_13_cast = conv(dilations = var_1675, groups = var_120, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = var_1673, 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_1679 = const()[name = tensor<string, []>("op_1679"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_1681 = const()[name = tensor<string, []>("op_1681"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_7_pad_type_0 = const()[name = tensor<string, []>("v_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_7_pad_0 = const()[name = tensor<string, []>("v_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2587584))), lut = tensor<fp16, [4]>([-0x1.8p-6, -0x1.7c4p-8, 0x1.878p-8, 0x1.8a4p-6]), name = tensor<string, []>("down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_7_cast = conv(dilations = var_1681, groups = var_120, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = var_1679, 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_1685_begin_0 = const()[name = tensor<string, []>("op_1685_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1685_end_0 = const()[name = tensor<string, []>("op_1685_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1685_end_mask_0 = const()[name = tensor<string, []>("op_1685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1685_cast = slice_by_index(begin = var_1685_begin_0, end = var_1685_end_0, end_mask = var_1685_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1685_cast")];
            tensor<int32, [4]> var_1689_begin_0 = const()[name = tensor<string, []>("op_1689_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1689_end_0 = const()[name = tensor<string, []>("op_1689_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_1689_end_mask_0 = const()[name = tensor<string, []>("op_1689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1689_cast = slice_by_index(begin = var_1689_begin_0, end = var_1689_end_0, end_mask = var_1689_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1689_cast")];
            tensor<int32, [4]> var_1693_begin_0 = const()[name = tensor<string, []>("op_1693_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1693_end_0 = const()[name = tensor<string, []>("op_1693_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_1693_end_mask_0 = const()[name = tensor<string, []>("op_1693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1693_cast = slice_by_index(begin = var_1693_begin_0, end = var_1693_end_0, end_mask = var_1693_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1693_cast")];
            tensor<int32, [4]> var_1697_begin_0 = const()[name = tensor<string, []>("op_1697_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1697_end_0 = const()[name = tensor<string, []>("op_1697_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_1697_end_mask_0 = const()[name = tensor<string, []>("op_1697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1697_cast = slice_by_index(begin = var_1697_begin_0, end = var_1697_end_0, end_mask = var_1697_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1697_cast")];
            tensor<int32, [4]> var_1701_begin_0 = const()[name = tensor<string, []>("op_1701_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1701_end_0 = const()[name = tensor<string, []>("op_1701_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_1701_end_mask_0 = const()[name = tensor<string, []>("op_1701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_1701_cast = slice_by_index(begin = var_1701_begin_0, end = var_1701_end_0, end_mask = var_1701_end_mask_0, x = q_7_cast)[name = tensor<string, []>("op_1701_cast")];
            tensor<int32, [4]> var_1704_begin_0 = const()[name = tensor<string, []>("op_1704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1704_end_0 = const()[name = tensor<string, []>("op_1704_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1704_end_mask_0 = const()[name = tensor<string, []>("op_1704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1704_cast = slice_by_index(begin = var_1704_begin_0, end = var_1704_end_0, end_mask = var_1704_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1704_cast")];
            tensor<int32, [4]> var_1705_begin_0 = const()[name = tensor<string, []>("op_1705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1705_end_0 = const()[name = tensor<string, []>("op_1705_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1705_end_mask_0 = const()[name = tensor<string, []>("op_1705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1705_cast = slice_by_index(begin = var_1705_begin_0, end = var_1705_end_0, end_mask = var_1705_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1705_cast")];
            tensor<int32, [4]> var_1706_begin_0 = const()[name = tensor<string, []>("op_1706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1706_end_0 = const()[name = tensor<string, []>("op_1706_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1706_end_mask_0 = const()[name = tensor<string, []>("op_1706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1706_cast = slice_by_index(begin = var_1706_begin_0, end = var_1706_end_0, end_mask = var_1706_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1706_cast")];
            tensor<int32, [4]> var_1707_begin_0 = const()[name = tensor<string, []>("op_1707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1707_end_0 = const()[name = tensor<string, []>("op_1707_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1707_end_mask_0 = const()[name = tensor<string, []>("op_1707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1707_cast = slice_by_index(begin = var_1707_begin_0, end = var_1707_end_0, end_mask = var_1707_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1707_cast")];
            tensor<int32, [4]> var_1708_begin_0 = const()[name = tensor<string, []>("op_1708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1708_end_0 = const()[name = tensor<string, []>("op_1708_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1708_end_mask_0 = const()[name = tensor<string, []>("op_1708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1708_cast = slice_by_index(begin = var_1708_begin_0, end = var_1708_end_0, end_mask = var_1708_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1708_cast")];
            tensor<int32, [4]> var_1709_begin_0 = const()[name = tensor<string, []>("op_1709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1709_end_0 = const()[name = tensor<string, []>("op_1709_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1709_end_mask_0 = const()[name = tensor<string, []>("op_1709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1709_cast = slice_by_index(begin = var_1709_begin_0, end = var_1709_end_0, end_mask = var_1709_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1709_cast")];
            tensor<int32, [4]> var_1710_begin_0 = const()[name = tensor<string, []>("op_1710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1710_end_0 = const()[name = tensor<string, []>("op_1710_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1710_end_mask_0 = const()[name = tensor<string, []>("op_1710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1710_cast = slice_by_index(begin = var_1710_begin_0, end = var_1710_end_0, end_mask = var_1710_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1710_cast")];
            tensor<int32, [4]> var_1711_begin_0 = const()[name = tensor<string, []>("op_1711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1711_end_0 = const()[name = tensor<string, []>("op_1711_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1711_end_mask_0 = const()[name = tensor<string, []>("op_1711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1711_cast = slice_by_index(begin = var_1711_begin_0, end = var_1711_end_0, end_mask = var_1711_end_mask_0, x = var_1685_cast)[name = tensor<string, []>("op_1711_cast")];
            tensor<int32, [4]> var_1712_begin_0 = const()[name = tensor<string, []>("op_1712_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1712_end_0 = const()[name = tensor<string, []>("op_1712_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1712_end_mask_0 = const()[name = tensor<string, []>("op_1712_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1712_cast = slice_by_index(begin = var_1712_begin_0, end = var_1712_end_0, end_mask = var_1712_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1712_cast")];
            tensor<int32, [4]> var_1713_begin_0 = const()[name = tensor<string, []>("op_1713_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1713_end_0 = const()[name = tensor<string, []>("op_1713_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1713_end_mask_0 = const()[name = tensor<string, []>("op_1713_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1713_cast = slice_by_index(begin = var_1713_begin_0, end = var_1713_end_0, end_mask = var_1713_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1713_cast")];
            tensor<int32, [4]> var_1714_begin_0 = const()[name = tensor<string, []>("op_1714_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1714_end_0 = const()[name = tensor<string, []>("op_1714_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1714_end_mask_0 = const()[name = tensor<string, []>("op_1714_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1714_cast = slice_by_index(begin = var_1714_begin_0, end = var_1714_end_0, end_mask = var_1714_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1714_cast")];
            tensor<int32, [4]> var_1715_begin_0 = const()[name = tensor<string, []>("op_1715_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1715_end_0 = const()[name = tensor<string, []>("op_1715_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1715_end_mask_0 = const()[name = tensor<string, []>("op_1715_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1715_cast = slice_by_index(begin = var_1715_begin_0, end = var_1715_end_0, end_mask = var_1715_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1715_cast")];
            tensor<int32, [4]> var_1716_begin_0 = const()[name = tensor<string, []>("op_1716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1716_end_0 = const()[name = tensor<string, []>("op_1716_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1716_end_mask_0 = const()[name = tensor<string, []>("op_1716_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1716_cast = slice_by_index(begin = var_1716_begin_0, end = var_1716_end_0, end_mask = var_1716_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1716_cast")];
            tensor<int32, [4]> var_1717_begin_0 = const()[name = tensor<string, []>("op_1717_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1717_end_0 = const()[name = tensor<string, []>("op_1717_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1717_end_mask_0 = const()[name = tensor<string, []>("op_1717_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1717_cast = slice_by_index(begin = var_1717_begin_0, end = var_1717_end_0, end_mask = var_1717_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1717_cast")];
            tensor<int32, [4]> var_1718_begin_0 = const()[name = tensor<string, []>("op_1718_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1718_end_0 = const()[name = tensor<string, []>("op_1718_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1718_end_mask_0 = const()[name = tensor<string, []>("op_1718_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1718_cast = slice_by_index(begin = var_1718_begin_0, end = var_1718_end_0, end_mask = var_1718_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1718_cast")];
            tensor<int32, [4]> var_1719_begin_0 = const()[name = tensor<string, []>("op_1719_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1719_end_0 = const()[name = tensor<string, []>("op_1719_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1719_end_mask_0 = const()[name = tensor<string, []>("op_1719_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1719_cast = slice_by_index(begin = var_1719_begin_0, end = var_1719_end_0, end_mask = var_1719_end_mask_0, x = var_1689_cast)[name = tensor<string, []>("op_1719_cast")];
            tensor<int32, [4]> var_1720_begin_0 = const()[name = tensor<string, []>("op_1720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1720_end_0 = const()[name = tensor<string, []>("op_1720_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1720_end_mask_0 = const()[name = tensor<string, []>("op_1720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1720_cast = slice_by_index(begin = var_1720_begin_0, end = var_1720_end_0, end_mask = var_1720_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1720_cast")];
            tensor<int32, [4]> var_1721_begin_0 = const()[name = tensor<string, []>("op_1721_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1721_end_0 = const()[name = tensor<string, []>("op_1721_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1721_end_mask_0 = const()[name = tensor<string, []>("op_1721_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1721_cast = slice_by_index(begin = var_1721_begin_0, end = var_1721_end_0, end_mask = var_1721_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1721_cast")];
            tensor<int32, [4]> var_1722_begin_0 = const()[name = tensor<string, []>("op_1722_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1722_end_0 = const()[name = tensor<string, []>("op_1722_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1722_end_mask_0 = const()[name = tensor<string, []>("op_1722_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1722_cast = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1722_cast")];
            tensor<int32, [4]> var_1723_begin_0 = const()[name = tensor<string, []>("op_1723_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1723_end_0 = const()[name = tensor<string, []>("op_1723_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1723_end_mask_0 = const()[name = tensor<string, []>("op_1723_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1723_cast = slice_by_index(begin = var_1723_begin_0, end = var_1723_end_0, end_mask = var_1723_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1723_cast")];
            tensor<int32, [4]> var_1724_begin_0 = const()[name = tensor<string, []>("op_1724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1724_end_0 = const()[name = tensor<string, []>("op_1724_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1724_end_mask_0 = const()[name = tensor<string, []>("op_1724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1724_cast = slice_by_index(begin = var_1724_begin_0, end = var_1724_end_0, end_mask = var_1724_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1724_cast")];
            tensor<int32, [4]> var_1725_begin_0 = const()[name = tensor<string, []>("op_1725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1725_end_0 = const()[name = tensor<string, []>("op_1725_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1725_end_mask_0 = const()[name = tensor<string, []>("op_1725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1725_cast = slice_by_index(begin = var_1725_begin_0, end = var_1725_end_0, end_mask = var_1725_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1725_cast")];
            tensor<int32, [4]> var_1726_begin_0 = const()[name = tensor<string, []>("op_1726_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1726_end_0 = const()[name = tensor<string, []>("op_1726_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1726_end_mask_0 = const()[name = tensor<string, []>("op_1726_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1726_cast = slice_by_index(begin = var_1726_begin_0, end = var_1726_end_0, end_mask = var_1726_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1726_cast")];
            tensor<int32, [4]> var_1727_begin_0 = const()[name = tensor<string, []>("op_1727_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1727_end_0 = const()[name = tensor<string, []>("op_1727_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1727_end_mask_0 = const()[name = tensor<string, []>("op_1727_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1727_cast = slice_by_index(begin = var_1727_begin_0, end = var_1727_end_0, end_mask = var_1727_end_mask_0, x = var_1693_cast)[name = tensor<string, []>("op_1727_cast")];
            tensor<int32, [4]> var_1728_begin_0 = const()[name = tensor<string, []>("op_1728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1728_end_0 = const()[name = tensor<string, []>("op_1728_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1728_end_mask_0 = const()[name = tensor<string, []>("op_1728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1728_cast = slice_by_index(begin = var_1728_begin_0, end = var_1728_end_0, end_mask = var_1728_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1728_cast")];
            tensor<int32, [4]> var_1729_begin_0 = const()[name = tensor<string, []>("op_1729_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1729_end_0 = const()[name = tensor<string, []>("op_1729_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1729_end_mask_0 = const()[name = tensor<string, []>("op_1729_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1729_cast = slice_by_index(begin = var_1729_begin_0, end = var_1729_end_0, end_mask = var_1729_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1729_cast")];
            tensor<int32, [4]> var_1730_begin_0 = const()[name = tensor<string, []>("op_1730_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1730_end_0 = const()[name = tensor<string, []>("op_1730_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1730_end_mask_0 = const()[name = tensor<string, []>("op_1730_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1730_cast = slice_by_index(begin = var_1730_begin_0, end = var_1730_end_0, end_mask = var_1730_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1730_cast")];
            tensor<int32, [4]> var_1731_begin_0 = const()[name = tensor<string, []>("op_1731_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1731_end_0 = const()[name = tensor<string, []>("op_1731_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1731_end_mask_0 = const()[name = tensor<string, []>("op_1731_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1731_cast = slice_by_index(begin = var_1731_begin_0, end = var_1731_end_0, end_mask = var_1731_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1731_cast")];
            tensor<int32, [4]> var_1732_begin_0 = const()[name = tensor<string, []>("op_1732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1732_end_0 = const()[name = tensor<string, []>("op_1732_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1732_end_mask_0 = const()[name = tensor<string, []>("op_1732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1732_cast = slice_by_index(begin = var_1732_begin_0, end = var_1732_end_0, end_mask = var_1732_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1732_cast")];
            tensor<int32, [4]> var_1733_begin_0 = const()[name = tensor<string, []>("op_1733_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1733_end_0 = const()[name = tensor<string, []>("op_1733_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1733_end_mask_0 = const()[name = tensor<string, []>("op_1733_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1733_cast = slice_by_index(begin = var_1733_begin_0, end = var_1733_end_0, end_mask = var_1733_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1733_cast")];
            tensor<int32, [4]> var_1734_begin_0 = const()[name = tensor<string, []>("op_1734_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1734_end_0 = const()[name = tensor<string, []>("op_1734_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1734_end_mask_0 = const()[name = tensor<string, []>("op_1734_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1734_cast = slice_by_index(begin = var_1734_begin_0, end = var_1734_end_0, end_mask = var_1734_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1734_cast")];
            tensor<int32, [4]> var_1735_begin_0 = const()[name = tensor<string, []>("op_1735_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1735_end_0 = const()[name = tensor<string, []>("op_1735_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1735_end_mask_0 = const()[name = tensor<string, []>("op_1735_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1735_cast = slice_by_index(begin = var_1735_begin_0, end = var_1735_end_0, end_mask = var_1735_end_mask_0, x = var_1697_cast)[name = tensor<string, []>("op_1735_cast")];
            tensor<int32, [4]> var_1736_begin_0 = const()[name = tensor<string, []>("op_1736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1736_end_0 = const()[name = tensor<string, []>("op_1736_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_1736_end_mask_0 = const()[name = tensor<string, []>("op_1736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1736_cast = slice_by_index(begin = var_1736_begin_0, end = var_1736_end_0, end_mask = var_1736_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1736_cast")];
            tensor<int32, [4]> var_1737_begin_0 = const()[name = tensor<string, []>("op_1737_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_1737_end_0 = const()[name = tensor<string, []>("op_1737_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_1737_end_mask_0 = const()[name = tensor<string, []>("op_1737_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1737_cast = slice_by_index(begin = var_1737_begin_0, end = var_1737_end_0, end_mask = var_1737_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1737_cast")];
            tensor<int32, [4]> var_1738_begin_0 = const()[name = tensor<string, []>("op_1738_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_1738_end_0 = const()[name = tensor<string, []>("op_1738_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_1738_end_mask_0 = const()[name = tensor<string, []>("op_1738_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1738_cast = slice_by_index(begin = var_1738_begin_0, end = var_1738_end_0, end_mask = var_1738_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1738_cast")];
            tensor<int32, [4]> var_1739_begin_0 = const()[name = tensor<string, []>("op_1739_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_1739_end_0 = const()[name = tensor<string, []>("op_1739_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_1739_end_mask_0 = const()[name = tensor<string, []>("op_1739_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1739_cast = slice_by_index(begin = var_1739_begin_0, end = var_1739_end_0, end_mask = var_1739_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1739_cast")];
            tensor<int32, [4]> var_1740_begin_0 = const()[name = tensor<string, []>("op_1740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_1740_end_0 = const()[name = tensor<string, []>("op_1740_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_1740_end_mask_0 = const()[name = tensor<string, []>("op_1740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1740_cast = slice_by_index(begin = var_1740_begin_0, end = var_1740_end_0, end_mask = var_1740_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1740_cast")];
            tensor<int32, [4]> var_1741_begin_0 = const()[name = tensor<string, []>("op_1741_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_1741_end_0 = const()[name = tensor<string, []>("op_1741_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_1741_end_mask_0 = const()[name = tensor<string, []>("op_1741_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1741_cast = slice_by_index(begin = var_1741_begin_0, end = var_1741_end_0, end_mask = var_1741_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1741_cast")];
            tensor<int32, [4]> var_1742_begin_0 = const()[name = tensor<string, []>("op_1742_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_1742_end_0 = const()[name = tensor<string, []>("op_1742_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_1742_end_mask_0 = const()[name = tensor<string, []>("op_1742_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1742_cast = slice_by_index(begin = var_1742_begin_0, end = var_1742_end_0, end_mask = var_1742_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1742_cast")];
            tensor<int32, [4]> var_1743_begin_0 = const()[name = tensor<string, []>("op_1743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_1743_end_0 = const()[name = tensor<string, []>("op_1743_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_1743_end_mask_0 = const()[name = tensor<string, []>("op_1743_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_1743_cast = slice_by_index(begin = var_1743_begin_0, end = var_1743_end_0, end_mask = var_1743_end_mask_0, x = var_1701_cast)[name = tensor<string, []>("op_1743_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_1748_begin_0 = const()[name = tensor<string, []>("op_1748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1748_end_0 = const()[name = tensor<string, []>("op_1748_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            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, 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, 64]> var_1748_cast = slice_by_index(begin = var_1748_begin_0, end = var_1748_end_0, end_mask = var_1748_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_1748_cast")];
            tensor<int32, [4]> var_1752_begin_0 = const()[name = tensor<string, []>("op_1752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1752_end_0 = const()[name = tensor<string, []>("op_1752_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            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, 77, 1, 64]> var_1752_cast = slice_by_index(begin = var_1752_begin_0, end = var_1752_end_0, end_mask = var_1752_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_1752_cast")];
            tensor<int32, [4]> var_1756_begin_0 = const()[name = tensor<string, []>("op_1756_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1756_end_0 = const()[name = tensor<string, []>("op_1756_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            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, 77, 1, 64]> var_1756_cast = slice_by_index(begin = var_1756_begin_0, end = var_1756_end_0, end_mask = var_1756_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_1756_cast")];
            tensor<int32, [4]> var_1760_begin_0 = const()[name = tensor<string, []>("op_1760_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1760_end_0 = const()[name = tensor<string, []>("op_1760_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            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, 77, 1, 64]> var_1760_cast = slice_by_index(begin = var_1760_begin_0, end = var_1760_end_0, end_mask = var_1760_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_1760_cast")];
            tensor<int32, [4]> var_1764_begin_0 = const()[name = tensor<string, []>("op_1764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1764_end_0 = const()[name = tensor<string, []>("op_1764_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            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, 77, 1, 64]> var_1764_cast = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = transpose_28)[name = tensor<string, []>("op_1764_cast")];
            tensor<int32, [4]> var_1766_begin_0 = const()[name = tensor<string, []>("op_1766_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1766_end_0 = const()[name = tensor<string, []>("op_1766_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_1766_end_mask_0 = const()[name = tensor<string, []>("op_1766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1766_cast = slice_by_index(begin = var_1766_begin_0, end = var_1766_end_0, end_mask = var_1766_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1766_cast")];
            tensor<int32, [4]> var_1770_begin_0 = const()[name = tensor<string, []>("op_1770_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1770_end_0 = const()[name = tensor<string, []>("op_1770_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_1770_end_mask_0 = const()[name = tensor<string, []>("op_1770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1770_cast = slice_by_index(begin = var_1770_begin_0, end = var_1770_end_0, end_mask = var_1770_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1770_cast")];
            tensor<int32, [4]> var_1774_begin_0 = const()[name = tensor<string, []>("op_1774_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1774_end_0 = const()[name = tensor<string, []>("op_1774_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_1774_end_mask_0 = const()[name = tensor<string, []>("op_1774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1774_cast = slice_by_index(begin = var_1774_begin_0, end = var_1774_end_0, end_mask = var_1774_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1774_cast")];
            tensor<int32, [4]> var_1778_begin_0 = const()[name = tensor<string, []>("op_1778_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1778_end_0 = const()[name = tensor<string, []>("op_1778_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_1778_end_mask_0 = const()[name = tensor<string, []>("op_1778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1778_cast = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1778_cast")];
            tensor<int32, [4]> var_1782_begin_0 = const()[name = tensor<string, []>("op_1782_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1782_end_0 = const()[name = tensor<string, []>("op_1782_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_1782_end_mask_0 = const()[name = tensor<string, []>("op_1782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_1782_cast = slice_by_index(begin = var_1782_begin_0, end = var_1782_end_0, end_mask = var_1782_end_mask_0, x = v_7_cast)[name = tensor<string, []>("op_1782_cast")];
            tensor<string, []> var_1786_equation_0 = const()[name = tensor<string, []>("op_1786_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1786_cast = einsum(equation = var_1786_equation_0, values = (var_1748_cast, var_1704_cast))[name = tensor<string, []>("op_1786_cast")];
            tensor<fp16, []> var_1787_to_fp16 = const()[name = tensor<string, []>("op_1787_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_241_cast = mul(x = var_1786_cast, y = var_1787_to_fp16)[name = tensor<string, []>("aw_chunk_241_cast")];
            tensor<string, []> var_1790_equation_0 = const()[name = tensor<string, []>("op_1790_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1790_cast = einsum(equation = var_1790_equation_0, values = (var_1748_cast, var_1705_cast))[name = tensor<string, []>("op_1790_cast")];
            tensor<fp16, []> var_1791_to_fp16 = const()[name = tensor<string, []>("op_1791_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_243_cast = mul(x = var_1790_cast, y = var_1791_to_fp16)[name = tensor<string, []>("aw_chunk_243_cast")];
            tensor<string, []> var_1794_equation_0 = const()[name = tensor<string, []>("op_1794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1794_cast = einsum(equation = var_1794_equation_0, values = (var_1748_cast, var_1706_cast))[name = tensor<string, []>("op_1794_cast")];
            tensor<fp16, []> var_1795_to_fp16 = const()[name = tensor<string, []>("op_1795_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_245_cast = mul(x = var_1794_cast, y = var_1795_to_fp16)[name = tensor<string, []>("aw_chunk_245_cast")];
            tensor<string, []> var_1798_equation_0 = const()[name = tensor<string, []>("op_1798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1798_cast = einsum(equation = var_1798_equation_0, values = (var_1748_cast, var_1707_cast))[name = tensor<string, []>("op_1798_cast")];
            tensor<fp16, []> var_1799_to_fp16 = const()[name = tensor<string, []>("op_1799_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_247_cast = mul(x = var_1798_cast, y = var_1799_to_fp16)[name = tensor<string, []>("aw_chunk_247_cast")];
            tensor<string, []> var_1802_equation_0 = const()[name = tensor<string, []>("op_1802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1802_cast = einsum(equation = var_1802_equation_0, values = (var_1748_cast, var_1708_cast))[name = tensor<string, []>("op_1802_cast")];
            tensor<fp16, []> var_1803_to_fp16 = const()[name = tensor<string, []>("op_1803_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_249_cast = mul(x = var_1802_cast, y = var_1803_to_fp16)[name = tensor<string, []>("aw_chunk_249_cast")];
            tensor<string, []> var_1806_equation_0 = const()[name = tensor<string, []>("op_1806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1806_cast = einsum(equation = var_1806_equation_0, values = (var_1748_cast, var_1709_cast))[name = tensor<string, []>("op_1806_cast")];
            tensor<fp16, []> var_1807_to_fp16 = const()[name = tensor<string, []>("op_1807_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_251_cast = mul(x = var_1806_cast, y = var_1807_to_fp16)[name = tensor<string, []>("aw_chunk_251_cast")];
            tensor<string, []> var_1810_equation_0 = const()[name = tensor<string, []>("op_1810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1810_cast = einsum(equation = var_1810_equation_0, values = (var_1748_cast, var_1710_cast))[name = tensor<string, []>("op_1810_cast")];
            tensor<fp16, []> var_1811_to_fp16 = const()[name = tensor<string, []>("op_1811_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_253_cast = mul(x = var_1810_cast, y = var_1811_to_fp16)[name = tensor<string, []>("aw_chunk_253_cast")];
            tensor<string, []> var_1814_equation_0 = const()[name = tensor<string, []>("op_1814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1814_cast = einsum(equation = var_1814_equation_0, values = (var_1748_cast, var_1711_cast))[name = tensor<string, []>("op_1814_cast")];
            tensor<fp16, []> var_1815_to_fp16 = const()[name = tensor<string, []>("op_1815_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_255_cast = mul(x = var_1814_cast, y = var_1815_to_fp16)[name = tensor<string, []>("aw_chunk_255_cast")];
            tensor<string, []> var_1818_equation_0 = const()[name = tensor<string, []>("op_1818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1818_cast = einsum(equation = var_1818_equation_0, values = (var_1752_cast, var_1712_cast))[name = tensor<string, []>("op_1818_cast")];
            tensor<fp16, []> var_1819_to_fp16 = const()[name = tensor<string, []>("op_1819_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_257_cast = mul(x = var_1818_cast, y = var_1819_to_fp16)[name = tensor<string, []>("aw_chunk_257_cast")];
            tensor<string, []> var_1822_equation_0 = const()[name = tensor<string, []>("op_1822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1822_cast = einsum(equation = var_1822_equation_0, values = (var_1752_cast, var_1713_cast))[name = tensor<string, []>("op_1822_cast")];
            tensor<fp16, []> var_1823_to_fp16 = const()[name = tensor<string, []>("op_1823_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_259_cast = mul(x = var_1822_cast, y = var_1823_to_fp16)[name = tensor<string, []>("aw_chunk_259_cast")];
            tensor<string, []> var_1826_equation_0 = const()[name = tensor<string, []>("op_1826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1826_cast = einsum(equation = var_1826_equation_0, values = (var_1752_cast, var_1714_cast))[name = tensor<string, []>("op_1826_cast")];
            tensor<fp16, []> var_1827_to_fp16 = const()[name = tensor<string, []>("op_1827_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_261_cast = mul(x = var_1826_cast, y = var_1827_to_fp16)[name = tensor<string, []>("aw_chunk_261_cast")];
            tensor<string, []> var_1830_equation_0 = const()[name = tensor<string, []>("op_1830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1830_cast = einsum(equation = var_1830_equation_0, values = (var_1752_cast, var_1715_cast))[name = tensor<string, []>("op_1830_cast")];
            tensor<fp16, []> var_1831_to_fp16 = const()[name = tensor<string, []>("op_1831_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_263_cast = mul(x = var_1830_cast, y = var_1831_to_fp16)[name = tensor<string, []>("aw_chunk_263_cast")];
            tensor<string, []> var_1834_equation_0 = const()[name = tensor<string, []>("op_1834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1834_cast = einsum(equation = var_1834_equation_0, values = (var_1752_cast, var_1716_cast))[name = tensor<string, []>("op_1834_cast")];
            tensor<fp16, []> var_1835_to_fp16 = const()[name = tensor<string, []>("op_1835_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_265_cast = mul(x = var_1834_cast, y = var_1835_to_fp16)[name = tensor<string, []>("aw_chunk_265_cast")];
            tensor<string, []> var_1838_equation_0 = const()[name = tensor<string, []>("op_1838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1838_cast = einsum(equation = var_1838_equation_0, values = (var_1752_cast, var_1717_cast))[name = tensor<string, []>("op_1838_cast")];
            tensor<fp16, []> var_1839_to_fp16 = const()[name = tensor<string, []>("op_1839_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_267_cast = mul(x = var_1838_cast, y = var_1839_to_fp16)[name = tensor<string, []>("aw_chunk_267_cast")];
            tensor<string, []> var_1842_equation_0 = const()[name = tensor<string, []>("op_1842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1842_cast = einsum(equation = var_1842_equation_0, values = (var_1752_cast, var_1718_cast))[name = tensor<string, []>("op_1842_cast")];
            tensor<fp16, []> var_1843_to_fp16 = const()[name = tensor<string, []>("op_1843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_269_cast = mul(x = var_1842_cast, y = var_1843_to_fp16)[name = tensor<string, []>("aw_chunk_269_cast")];
            tensor<string, []> var_1846_equation_0 = const()[name = tensor<string, []>("op_1846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1846_cast = einsum(equation = var_1846_equation_0, values = (var_1752_cast, var_1719_cast))[name = tensor<string, []>("op_1846_cast")];
            tensor<fp16, []> var_1847_to_fp16 = const()[name = tensor<string, []>("op_1847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_271_cast = mul(x = var_1846_cast, y = var_1847_to_fp16)[name = tensor<string, []>("aw_chunk_271_cast")];
            tensor<string, []> var_1850_equation_0 = const()[name = tensor<string, []>("op_1850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1850_cast = einsum(equation = var_1850_equation_0, values = (var_1756_cast, var_1720_cast))[name = tensor<string, []>("op_1850_cast")];
            tensor<fp16, []> var_1851_to_fp16 = const()[name = tensor<string, []>("op_1851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_273_cast = mul(x = var_1850_cast, y = var_1851_to_fp16)[name = tensor<string, []>("aw_chunk_273_cast")];
            tensor<string, []> var_1854_equation_0 = const()[name = tensor<string, []>("op_1854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1854_cast = einsum(equation = var_1854_equation_0, values = (var_1756_cast, var_1721_cast))[name = tensor<string, []>("op_1854_cast")];
            tensor<fp16, []> var_1855_to_fp16 = const()[name = tensor<string, []>("op_1855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_275_cast = mul(x = var_1854_cast, y = var_1855_to_fp16)[name = tensor<string, []>("aw_chunk_275_cast")];
            tensor<string, []> var_1858_equation_0 = const()[name = tensor<string, []>("op_1858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1858_cast = einsum(equation = var_1858_equation_0, values = (var_1756_cast, var_1722_cast))[name = tensor<string, []>("op_1858_cast")];
            tensor<fp16, []> var_1859_to_fp16 = const()[name = tensor<string, []>("op_1859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_277_cast = mul(x = var_1858_cast, y = var_1859_to_fp16)[name = tensor<string, []>("aw_chunk_277_cast")];
            tensor<string, []> var_1862_equation_0 = const()[name = tensor<string, []>("op_1862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1862_cast = einsum(equation = var_1862_equation_0, values = (var_1756_cast, var_1723_cast))[name = tensor<string, []>("op_1862_cast")];
            tensor<fp16, []> var_1863_to_fp16 = const()[name = tensor<string, []>("op_1863_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_279_cast = mul(x = var_1862_cast, y = var_1863_to_fp16)[name = tensor<string, []>("aw_chunk_279_cast")];
            tensor<string, []> var_1866_equation_0 = const()[name = tensor<string, []>("op_1866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1866_cast = einsum(equation = var_1866_equation_0, values = (var_1756_cast, var_1724_cast))[name = tensor<string, []>("op_1866_cast")];
            tensor<fp16, []> var_1867_to_fp16 = const()[name = tensor<string, []>("op_1867_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_281_cast = mul(x = var_1866_cast, y = var_1867_to_fp16)[name = tensor<string, []>("aw_chunk_281_cast")];
            tensor<string, []> var_1870_equation_0 = const()[name = tensor<string, []>("op_1870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1870_cast = einsum(equation = var_1870_equation_0, values = (var_1756_cast, var_1725_cast))[name = tensor<string, []>("op_1870_cast")];
            tensor<fp16, []> var_1871_to_fp16 = const()[name = tensor<string, []>("op_1871_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_283_cast = mul(x = var_1870_cast, y = var_1871_to_fp16)[name = tensor<string, []>("aw_chunk_283_cast")];
            tensor<string, []> var_1874_equation_0 = const()[name = tensor<string, []>("op_1874_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1874_cast = einsum(equation = var_1874_equation_0, values = (var_1756_cast, var_1726_cast))[name = tensor<string, []>("op_1874_cast")];
            tensor<fp16, []> var_1875_to_fp16 = const()[name = tensor<string, []>("op_1875_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_285_cast = mul(x = var_1874_cast, y = var_1875_to_fp16)[name = tensor<string, []>("aw_chunk_285_cast")];
            tensor<string, []> var_1878_equation_0 = const()[name = tensor<string, []>("op_1878_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1878_cast = einsum(equation = var_1878_equation_0, values = (var_1756_cast, var_1727_cast))[name = tensor<string, []>("op_1878_cast")];
            tensor<fp16, []> var_1879_to_fp16 = const()[name = tensor<string, []>("op_1879_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_287_cast = mul(x = var_1878_cast, y = var_1879_to_fp16)[name = tensor<string, []>("aw_chunk_287_cast")];
            tensor<string, []> var_1882_equation_0 = const()[name = tensor<string, []>("op_1882_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1882_cast = einsum(equation = var_1882_equation_0, values = (var_1760_cast, var_1728_cast))[name = tensor<string, []>("op_1882_cast")];
            tensor<fp16, []> var_1883_to_fp16 = const()[name = tensor<string, []>("op_1883_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_289_cast = mul(x = var_1882_cast, y = var_1883_to_fp16)[name = tensor<string, []>("aw_chunk_289_cast")];
            tensor<string, []> var_1886_equation_0 = const()[name = tensor<string, []>("op_1886_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1886_cast = einsum(equation = var_1886_equation_0, values = (var_1760_cast, var_1729_cast))[name = tensor<string, []>("op_1886_cast")];
            tensor<fp16, []> var_1887_to_fp16 = const()[name = tensor<string, []>("op_1887_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_291_cast = mul(x = var_1886_cast, y = var_1887_to_fp16)[name = tensor<string, []>("aw_chunk_291_cast")];
            tensor<string, []> var_1890_equation_0 = const()[name = tensor<string, []>("op_1890_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1890_cast = einsum(equation = var_1890_equation_0, values = (var_1760_cast, var_1730_cast))[name = tensor<string, []>("op_1890_cast")];
            tensor<fp16, []> var_1891_to_fp16 = const()[name = tensor<string, []>("op_1891_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_293_cast = mul(x = var_1890_cast, y = var_1891_to_fp16)[name = tensor<string, []>("aw_chunk_293_cast")];
            tensor<string, []> var_1894_equation_0 = const()[name = tensor<string, []>("op_1894_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1894_cast = einsum(equation = var_1894_equation_0, values = (var_1760_cast, var_1731_cast))[name = tensor<string, []>("op_1894_cast")];
            tensor<fp16, []> var_1895_to_fp16 = const()[name = tensor<string, []>("op_1895_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_295_cast = mul(x = var_1894_cast, y = var_1895_to_fp16)[name = tensor<string, []>("aw_chunk_295_cast")];
            tensor<string, []> var_1898_equation_0 = const()[name = tensor<string, []>("op_1898_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1898_cast = einsum(equation = var_1898_equation_0, values = (var_1760_cast, var_1732_cast))[name = tensor<string, []>("op_1898_cast")];
            tensor<fp16, []> var_1899_to_fp16 = const()[name = tensor<string, []>("op_1899_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_297_cast = mul(x = var_1898_cast, y = var_1899_to_fp16)[name = tensor<string, []>("aw_chunk_297_cast")];
            tensor<string, []> var_1902_equation_0 = const()[name = tensor<string, []>("op_1902_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1902_cast = einsum(equation = var_1902_equation_0, values = (var_1760_cast, var_1733_cast))[name = tensor<string, []>("op_1902_cast")];
            tensor<fp16, []> var_1903_to_fp16 = const()[name = tensor<string, []>("op_1903_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_299_cast = mul(x = var_1902_cast, y = var_1903_to_fp16)[name = tensor<string, []>("aw_chunk_299_cast")];
            tensor<string, []> var_1906_equation_0 = const()[name = tensor<string, []>("op_1906_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1906_cast = einsum(equation = var_1906_equation_0, values = (var_1760_cast, var_1734_cast))[name = tensor<string, []>("op_1906_cast")];
            tensor<fp16, []> var_1907_to_fp16 = const()[name = tensor<string, []>("op_1907_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_301_cast = mul(x = var_1906_cast, y = var_1907_to_fp16)[name = tensor<string, []>("aw_chunk_301_cast")];
            tensor<string, []> var_1910_equation_0 = const()[name = tensor<string, []>("op_1910_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1910_cast = einsum(equation = var_1910_equation_0, values = (var_1760_cast, var_1735_cast))[name = tensor<string, []>("op_1910_cast")];
            tensor<fp16, []> var_1911_to_fp16 = const()[name = tensor<string, []>("op_1911_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_303_cast = mul(x = var_1910_cast, y = var_1911_to_fp16)[name = tensor<string, []>("aw_chunk_303_cast")];
            tensor<string, []> var_1914_equation_0 = const()[name = tensor<string, []>("op_1914_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1914_cast = einsum(equation = var_1914_equation_0, values = (var_1764_cast, var_1736_cast))[name = tensor<string, []>("op_1914_cast")];
            tensor<fp16, []> var_1915_to_fp16 = const()[name = tensor<string, []>("op_1915_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_305_cast = mul(x = var_1914_cast, y = var_1915_to_fp16)[name = tensor<string, []>("aw_chunk_305_cast")];
            tensor<string, []> var_1918_equation_0 = const()[name = tensor<string, []>("op_1918_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1918_cast = einsum(equation = var_1918_equation_0, values = (var_1764_cast, var_1737_cast))[name = tensor<string, []>("op_1918_cast")];
            tensor<fp16, []> var_1919_to_fp16 = const()[name = tensor<string, []>("op_1919_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_307_cast = mul(x = var_1918_cast, y = var_1919_to_fp16)[name = tensor<string, []>("aw_chunk_307_cast")];
            tensor<string, []> var_1922_equation_0 = const()[name = tensor<string, []>("op_1922_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1922_cast = einsum(equation = var_1922_equation_0, values = (var_1764_cast, var_1738_cast))[name = tensor<string, []>("op_1922_cast")];
            tensor<fp16, []> var_1923_to_fp16 = const()[name = tensor<string, []>("op_1923_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_309_cast = mul(x = var_1922_cast, y = var_1923_to_fp16)[name = tensor<string, []>("aw_chunk_309_cast")];
            tensor<string, []> var_1926_equation_0 = const()[name = tensor<string, []>("op_1926_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1926_cast = einsum(equation = var_1926_equation_0, values = (var_1764_cast, var_1739_cast))[name = tensor<string, []>("op_1926_cast")];
            tensor<fp16, []> var_1927_to_fp16 = const()[name = tensor<string, []>("op_1927_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_311_cast = mul(x = var_1926_cast, y = var_1927_to_fp16)[name = tensor<string, []>("aw_chunk_311_cast")];
            tensor<string, []> var_1930_equation_0 = const()[name = tensor<string, []>("op_1930_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1930_cast = einsum(equation = var_1930_equation_0, values = (var_1764_cast, var_1740_cast))[name = tensor<string, []>("op_1930_cast")];
            tensor<fp16, []> var_1931_to_fp16 = const()[name = tensor<string, []>("op_1931_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_313_cast = mul(x = var_1930_cast, y = var_1931_to_fp16)[name = tensor<string, []>("aw_chunk_313_cast")];
            tensor<string, []> var_1934_equation_0 = const()[name = tensor<string, []>("op_1934_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1934_cast = einsum(equation = var_1934_equation_0, values = (var_1764_cast, var_1741_cast))[name = tensor<string, []>("op_1934_cast")];
            tensor<fp16, []> var_1935_to_fp16 = const()[name = tensor<string, []>("op_1935_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_315_cast = mul(x = var_1934_cast, y = var_1935_to_fp16)[name = tensor<string, []>("aw_chunk_315_cast")];
            tensor<string, []> var_1938_equation_0 = const()[name = tensor<string, []>("op_1938_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1938_cast = einsum(equation = var_1938_equation_0, values = (var_1764_cast, var_1742_cast))[name = tensor<string, []>("op_1938_cast")];
            tensor<fp16, []> var_1939_to_fp16 = const()[name = tensor<string, []>("op_1939_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_317_cast = mul(x = var_1938_cast, y = var_1939_to_fp16)[name = tensor<string, []>("aw_chunk_317_cast")];
            tensor<string, []> var_1942_equation_0 = const()[name = tensor<string, []>("op_1942_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_1942_cast = einsum(equation = var_1942_equation_0, values = (var_1764_cast, var_1743_cast))[name = tensor<string, []>("op_1942_cast")];
            tensor<fp16, []> var_1943_to_fp16 = const()[name = tensor<string, []>("op_1943_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_319_cast = mul(x = var_1942_cast, y = var_1943_to_fp16)[name = tensor<string, []>("aw_chunk_319_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1945_cast = softmax(axis = var_120, x = aw_chunk_241_cast)[name = tensor<string, []>("op_1945_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1946_cast = softmax(axis = var_120, x = aw_chunk_243_cast)[name = tensor<string, []>("op_1946_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1947_cast = softmax(axis = var_120, x = aw_chunk_245_cast)[name = tensor<string, []>("op_1947_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1948_cast = softmax(axis = var_120, x = aw_chunk_247_cast)[name = tensor<string, []>("op_1948_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1949_cast = softmax(axis = var_120, x = aw_chunk_249_cast)[name = tensor<string, []>("op_1949_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1950_cast = softmax(axis = var_120, x = aw_chunk_251_cast)[name = tensor<string, []>("op_1950_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1951_cast = softmax(axis = var_120, x = aw_chunk_253_cast)[name = tensor<string, []>("op_1951_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1952_cast = softmax(axis = var_120, x = aw_chunk_255_cast)[name = tensor<string, []>("op_1952_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1953_cast = softmax(axis = var_120, x = aw_chunk_257_cast)[name = tensor<string, []>("op_1953_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1954_cast = softmax(axis = var_120, x = aw_chunk_259_cast)[name = tensor<string, []>("op_1954_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1955_cast = softmax(axis = var_120, x = aw_chunk_261_cast)[name = tensor<string, []>("op_1955_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1956_cast = softmax(axis = var_120, x = aw_chunk_263_cast)[name = tensor<string, []>("op_1956_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1957_cast = softmax(axis = var_120, x = aw_chunk_265_cast)[name = tensor<string, []>("op_1957_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1958_cast = softmax(axis = var_120, x = aw_chunk_267_cast)[name = tensor<string, []>("op_1958_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1959_cast = softmax(axis = var_120, x = aw_chunk_269_cast)[name = tensor<string, []>("op_1959_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1960_cast = softmax(axis = var_120, x = aw_chunk_271_cast)[name = tensor<string, []>("op_1960_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1961_cast = softmax(axis = var_120, x = aw_chunk_273_cast)[name = tensor<string, []>("op_1961_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1962_cast = softmax(axis = var_120, x = aw_chunk_275_cast)[name = tensor<string, []>("op_1962_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1963_cast = softmax(axis = var_120, x = aw_chunk_277_cast)[name = tensor<string, []>("op_1963_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1964_cast = softmax(axis = var_120, x = aw_chunk_279_cast)[name = tensor<string, []>("op_1964_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1965_cast = softmax(axis = var_120, x = aw_chunk_281_cast)[name = tensor<string, []>("op_1965_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1966_cast = softmax(axis = var_120, x = aw_chunk_283_cast)[name = tensor<string, []>("op_1966_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1967_cast = softmax(axis = var_120, x = aw_chunk_285_cast)[name = tensor<string, []>("op_1967_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1968_cast = softmax(axis = var_120, x = aw_chunk_287_cast)[name = tensor<string, []>("op_1968_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1969_cast = softmax(axis = var_120, x = aw_chunk_289_cast)[name = tensor<string, []>("op_1969_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1970_cast = softmax(axis = var_120, x = aw_chunk_291_cast)[name = tensor<string, []>("op_1970_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1971_cast = softmax(axis = var_120, x = aw_chunk_293_cast)[name = tensor<string, []>("op_1971_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1972_cast = softmax(axis = var_120, x = aw_chunk_295_cast)[name = tensor<string, []>("op_1972_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1973_cast = softmax(axis = var_120, x = aw_chunk_297_cast)[name = tensor<string, []>("op_1973_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1974_cast = softmax(axis = var_120, x = aw_chunk_299_cast)[name = tensor<string, []>("op_1974_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1975_cast = softmax(axis = var_120, x = aw_chunk_301_cast)[name = tensor<string, []>("op_1975_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1976_cast = softmax(axis = var_120, x = aw_chunk_303_cast)[name = tensor<string, []>("op_1976_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1977_cast = softmax(axis = var_120, x = aw_chunk_305_cast)[name = tensor<string, []>("op_1977_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1978_cast = softmax(axis = var_120, x = aw_chunk_307_cast)[name = tensor<string, []>("op_1978_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1979_cast = softmax(axis = var_120, x = aw_chunk_309_cast)[name = tensor<string, []>("op_1979_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1980_cast = softmax(axis = var_120, x = aw_chunk_311_cast)[name = tensor<string, []>("op_1980_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1981_cast = softmax(axis = var_120, x = aw_chunk_313_cast)[name = tensor<string, []>("op_1981_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1982_cast = softmax(axis = var_120, x = aw_chunk_315_cast)[name = tensor<string, []>("op_1982_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1983_cast = softmax(axis = var_120, x = aw_chunk_317_cast)[name = tensor<string, []>("op_1983_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_1984_cast = softmax(axis = var_120, x = aw_chunk_319_cast)[name = tensor<string, []>("op_1984_cast")];
            tensor<string, []> var_1986_equation_0 = const()[name = tensor<string, []>("op_1986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1986_cast = einsum(equation = var_1986_equation_0, values = (var_1766_cast, var_1945_cast))[name = tensor<string, []>("op_1986_cast")];
            tensor<string, []> var_1988_equation_0 = const()[name = tensor<string, []>("op_1988_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1988_cast = einsum(equation = var_1988_equation_0, values = (var_1766_cast, var_1946_cast))[name = tensor<string, []>("op_1988_cast")];
            tensor<string, []> var_1990_equation_0 = const()[name = tensor<string, []>("op_1990_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1990_cast = einsum(equation = var_1990_equation_0, values = (var_1766_cast, var_1947_cast))[name = tensor<string, []>("op_1990_cast")];
            tensor<string, []> var_1992_equation_0 = const()[name = tensor<string, []>("op_1992_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1992_cast = einsum(equation = var_1992_equation_0, values = (var_1766_cast, var_1948_cast))[name = tensor<string, []>("op_1992_cast")];
            tensor<string, []> var_1994_equation_0 = const()[name = tensor<string, []>("op_1994_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1994_cast = einsum(equation = var_1994_equation_0, values = (var_1766_cast, var_1949_cast))[name = tensor<string, []>("op_1994_cast")];
            tensor<string, []> var_1996_equation_0 = const()[name = tensor<string, []>("op_1996_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1996_cast = einsum(equation = var_1996_equation_0, values = (var_1766_cast, var_1950_cast))[name = tensor<string, []>("op_1996_cast")];
            tensor<string, []> var_1998_equation_0 = const()[name = tensor<string, []>("op_1998_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_1998_cast = einsum(equation = var_1998_equation_0, values = (var_1766_cast, var_1951_cast))[name = tensor<string, []>("op_1998_cast")];
            tensor<string, []> var_2000_equation_0 = const()[name = tensor<string, []>("op_2000_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2000_cast = einsum(equation = var_2000_equation_0, values = (var_1766_cast, var_1952_cast))[name = tensor<string, []>("op_2000_cast")];
            tensor<string, []> var_2002_equation_0 = const()[name = tensor<string, []>("op_2002_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2002_cast = einsum(equation = var_2002_equation_0, values = (var_1770_cast, var_1953_cast))[name = tensor<string, []>("op_2002_cast")];
            tensor<string, []> var_2004_equation_0 = const()[name = tensor<string, []>("op_2004_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2004_cast = einsum(equation = var_2004_equation_0, values = (var_1770_cast, var_1954_cast))[name = tensor<string, []>("op_2004_cast")];
            tensor<string, []> var_2006_equation_0 = const()[name = tensor<string, []>("op_2006_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2006_cast = einsum(equation = var_2006_equation_0, values = (var_1770_cast, var_1955_cast))[name = tensor<string, []>("op_2006_cast")];
            tensor<string, []> var_2008_equation_0 = const()[name = tensor<string, []>("op_2008_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2008_cast = einsum(equation = var_2008_equation_0, values = (var_1770_cast, var_1956_cast))[name = tensor<string, []>("op_2008_cast")];
            tensor<string, []> var_2010_equation_0 = const()[name = tensor<string, []>("op_2010_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2010_cast = einsum(equation = var_2010_equation_0, values = (var_1770_cast, var_1957_cast))[name = tensor<string, []>("op_2010_cast")];
            tensor<string, []> var_2012_equation_0 = const()[name = tensor<string, []>("op_2012_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2012_cast = einsum(equation = var_2012_equation_0, values = (var_1770_cast, var_1958_cast))[name = tensor<string, []>("op_2012_cast")];
            tensor<string, []> var_2014_equation_0 = const()[name = tensor<string, []>("op_2014_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2014_cast = einsum(equation = var_2014_equation_0, values = (var_1770_cast, var_1959_cast))[name = tensor<string, []>("op_2014_cast")];
            tensor<string, []> var_2016_equation_0 = const()[name = tensor<string, []>("op_2016_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2016_cast = einsum(equation = var_2016_equation_0, values = (var_1770_cast, var_1960_cast))[name = tensor<string, []>("op_2016_cast")];
            tensor<string, []> var_2018_equation_0 = const()[name = tensor<string, []>("op_2018_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2018_cast = einsum(equation = var_2018_equation_0, values = (var_1774_cast, var_1961_cast))[name = tensor<string, []>("op_2018_cast")];
            tensor<string, []> var_2020_equation_0 = const()[name = tensor<string, []>("op_2020_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2020_cast = einsum(equation = var_2020_equation_0, values = (var_1774_cast, var_1962_cast))[name = tensor<string, []>("op_2020_cast")];
            tensor<string, []> var_2022_equation_0 = const()[name = tensor<string, []>("op_2022_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2022_cast = einsum(equation = var_2022_equation_0, values = (var_1774_cast, var_1963_cast))[name = tensor<string, []>("op_2022_cast")];
            tensor<string, []> var_2024_equation_0 = const()[name = tensor<string, []>("op_2024_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2024_cast = einsum(equation = var_2024_equation_0, values = (var_1774_cast, var_1964_cast))[name = tensor<string, []>("op_2024_cast")];
            tensor<string, []> var_2026_equation_0 = const()[name = tensor<string, []>("op_2026_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2026_cast = einsum(equation = var_2026_equation_0, values = (var_1774_cast, var_1965_cast))[name = tensor<string, []>("op_2026_cast")];
            tensor<string, []> var_2028_equation_0 = const()[name = tensor<string, []>("op_2028_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2028_cast = einsum(equation = var_2028_equation_0, values = (var_1774_cast, var_1966_cast))[name = tensor<string, []>("op_2028_cast")];
            tensor<string, []> var_2030_equation_0 = const()[name = tensor<string, []>("op_2030_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2030_cast = einsum(equation = var_2030_equation_0, values = (var_1774_cast, var_1967_cast))[name = tensor<string, []>("op_2030_cast")];
            tensor<string, []> var_2032_equation_0 = const()[name = tensor<string, []>("op_2032_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2032_cast = einsum(equation = var_2032_equation_0, values = (var_1774_cast, var_1968_cast))[name = tensor<string, []>("op_2032_cast")];
            tensor<string, []> var_2034_equation_0 = const()[name = tensor<string, []>("op_2034_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2034_cast = einsum(equation = var_2034_equation_0, values = (var_1778_cast, var_1969_cast))[name = tensor<string, []>("op_2034_cast")];
            tensor<string, []> var_2036_equation_0 = const()[name = tensor<string, []>("op_2036_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2036_cast = einsum(equation = var_2036_equation_0, values = (var_1778_cast, var_1970_cast))[name = tensor<string, []>("op_2036_cast")];
            tensor<string, []> var_2038_equation_0 = const()[name = tensor<string, []>("op_2038_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2038_cast = einsum(equation = var_2038_equation_0, values = (var_1778_cast, var_1971_cast))[name = tensor<string, []>("op_2038_cast")];
            tensor<string, []> var_2040_equation_0 = const()[name = tensor<string, []>("op_2040_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2040_cast = einsum(equation = var_2040_equation_0, values = (var_1778_cast, var_1972_cast))[name = tensor<string, []>("op_2040_cast")];
            tensor<string, []> var_2042_equation_0 = const()[name = tensor<string, []>("op_2042_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2042_cast = einsum(equation = var_2042_equation_0, values = (var_1778_cast, var_1973_cast))[name = tensor<string, []>("op_2042_cast")];
            tensor<string, []> var_2044_equation_0 = const()[name = tensor<string, []>("op_2044_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2044_cast = einsum(equation = var_2044_equation_0, values = (var_1778_cast, var_1974_cast))[name = tensor<string, []>("op_2044_cast")];
            tensor<string, []> var_2046_equation_0 = const()[name = tensor<string, []>("op_2046_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2046_cast = einsum(equation = var_2046_equation_0, values = (var_1778_cast, var_1975_cast))[name = tensor<string, []>("op_2046_cast")];
            tensor<string, []> var_2048_equation_0 = const()[name = tensor<string, []>("op_2048_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2048_cast = einsum(equation = var_2048_equation_0, values = (var_1778_cast, var_1976_cast))[name = tensor<string, []>("op_2048_cast")];
            tensor<string, []> var_2050_equation_0 = const()[name = tensor<string, []>("op_2050_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2050_cast = einsum(equation = var_2050_equation_0, values = (var_1782_cast, var_1977_cast))[name = tensor<string, []>("op_2050_cast")];
            tensor<string, []> var_2052_equation_0 = const()[name = tensor<string, []>("op_2052_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2052_cast = einsum(equation = var_2052_equation_0, values = (var_1782_cast, var_1978_cast))[name = tensor<string, []>("op_2052_cast")];
            tensor<string, []> var_2054_equation_0 = const()[name = tensor<string, []>("op_2054_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2054_cast = einsum(equation = var_2054_equation_0, values = (var_1782_cast, var_1979_cast))[name = tensor<string, []>("op_2054_cast")];
            tensor<string, []> var_2056_equation_0 = const()[name = tensor<string, []>("op_2056_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2056_cast = einsum(equation = var_2056_equation_0, values = (var_1782_cast, var_1980_cast))[name = tensor<string, []>("op_2056_cast")];
            tensor<string, []> var_2058_equation_0 = const()[name = tensor<string, []>("op_2058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2058_cast = einsum(equation = var_2058_equation_0, values = (var_1782_cast, var_1981_cast))[name = tensor<string, []>("op_2058_cast")];
            tensor<string, []> var_2060_equation_0 = const()[name = tensor<string, []>("op_2060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2060_cast = einsum(equation = var_2060_equation_0, values = (var_1782_cast, var_1982_cast))[name = tensor<string, []>("op_2060_cast")];
            tensor<string, []> var_2062_equation_0 = const()[name = tensor<string, []>("op_2062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2062_cast = einsum(equation = var_2062_equation_0, values = (var_1782_cast, var_1983_cast))[name = tensor<string, []>("op_2062_cast")];
            tensor<string, []> var_2064_equation_0 = const()[name = tensor<string, []>("op_2064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2064_cast = einsum(equation = var_2064_equation_0, values = (var_1782_cast, var_1984_cast))[name = tensor<string, []>("op_2064_cast")];
            tensor<bool, []> var_2066_interleave_0 = const()[name = tensor<string, []>("op_2066_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_2066_cast = concat(axis = var_95, interleave = var_2066_interleave_0, values = (var_1986_cast, var_1988_cast, var_1990_cast, var_1992_cast, var_1994_cast, var_1996_cast, var_1998_cast, var_2000_cast))[name = tensor<string, []>("op_2066_cast")];
            tensor<bool, []> var_2068_interleave_0 = const()[name = tensor<string, []>("op_2068_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_2068_cast = concat(axis = var_95, interleave = var_2068_interleave_0, values = (var_2002_cast, var_2004_cast, var_2006_cast, var_2008_cast, var_2010_cast, var_2012_cast, var_2014_cast, var_2016_cast))[name = tensor<string, []>("op_2068_cast")];
            tensor<bool, []> var_2070_interleave_0 = const()[name = tensor<string, []>("op_2070_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_2070_cast = concat(axis = var_95, interleave = var_2070_interleave_0, values = (var_2018_cast, var_2020_cast, var_2022_cast, var_2024_cast, var_2026_cast, var_2028_cast, var_2030_cast, var_2032_cast))[name = tensor<string, []>("op_2070_cast")];
            tensor<bool, []> var_2072_interleave_0 = const()[name = tensor<string, []>("op_2072_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_2072_cast = concat(axis = var_95, interleave = var_2072_interleave_0, values = (var_2034_cast, var_2036_cast, var_2038_cast, var_2040_cast, var_2042_cast, var_2044_cast, var_2046_cast, var_2048_cast))[name = tensor<string, []>("op_2072_cast")];
            tensor<bool, []> var_2074_interleave_0 = const()[name = tensor<string, []>("op_2074_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_2074_cast = concat(axis = var_95, interleave = var_2074_interleave_0, values = (var_2050_cast, var_2052_cast, var_2054_cast, var_2056_cast, var_2058_cast, var_2060_cast, var_2062_cast, var_2064_cast))[name = tensor<string, []>("op_2074_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_120, interleave = input_53_interleave_0, values = (var_2066_cast, var_2068_cast, var_2070_cast, var_2072_cast, var_2074_cast))[name = tensor<string, []>("input_53_cast")];
            tensor<int32, [2]> var_2080 = const()[name = tensor<string, []>("op_2080"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2082 = const()[name = tensor<string, []>("op_2082"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2084_pad_type_0 = const()[name = tensor<string, []>("op_2084_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2084_pad_0 = const()[name = tensor<string, []>("op_2084_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2669568))), lut = tensor<fp16, [4]>([-0x1.8c8p-6, -0x1.cfp-10, 0x1.144p-9, 0x1.8b4p-6]), 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, []>(2695232)))];
            tensor<fp16, [2, 320, 1, 4096]> var_2084_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2082, groups = var_120, pad = var_2084_pad_0, pad_type = var_2084_pad_type_0, strides = var_2080, 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_2084_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_11_cast = add(x = var_2084_cast, y = inputs_9_cast)[name = tensor<string, []>("inputs_11_cast")];
            tensor<int32, [1]> var_2088 = const()[name = tensor<string, []>("op_2088"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_11_cast = reduce_mean(axes = var_2088, keep_dims = var_115, 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_2092 = const()[name = tensor<string, []>("op_2092"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_2093_cast = reduce_mean(axes = var_2092, keep_dims = var_115, x = zero_mean_sq_11_cast)[name = tensor<string, []>("op_2093_cast")];
            tensor<fp16, []> var_2094_to_fp16 = const()[name = tensor<string, []>("op_2094_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_2095_cast = add(x = var_2093_cast, y = var_2094_to_fp16)[name = tensor<string, []>("op_2095_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_2095_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_2099_to_fp16 = const()[name = tensor<string, []>("op_2099_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2695936)))];
            tensor<fp16, [2, 320, 1, 4096]> var_2100_cast = add(x = out_11_cast, y = var_2099_to_fp16)[name = tensor<string, []>("op_2100_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_2102_to_fp16 = const()[name = tensor<string, []>("op_2102_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2696640)))];
            tensor<fp16, [2, 320, 1, 4096]> input_55_cast = mul(x = var_2100_cast, y = var_2102_to_fp16)[name = tensor<string, []>("input_55_cast")];
            tensor<int32, [2]> var_2110 = const()[name = tensor<string, []>("op_2110"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2112 = const()[name = tensor<string, []>("op_2112"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2114_pad_type_0 = const()[name = tensor<string, []>("op_2114_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2114_pad_0 = const()[name = tensor<string, []>("op_2114_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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2697344))), lut = tensor<fp16, [4]>([-0x1.7ap-4, -0x1.c34p-6, 0x1.b08p-6, 0x1.754p-4]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2902208))), lut = tensor<fp16, [4]>([0x1.984p-5, -0x1.b3cp-5, 0x1.234p-8, 0x1.8ap-4]), 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_2114_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_2112, groups = var_120, pad = var_2114_pad_0, pad_type = var_2114_pad_type_0, strides = var_2110, 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_2114_cast")];
            tensor<int32, [2]> var_2115_split_sizes_0 = const()[name = tensor<string, []>("op_2115_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_2115_axis_0 = const()[name = tensor<string, []>("op_2115_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_2115_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_2115_cast_1 = split(axis = var_2115_axis_0, split_sizes = var_2115_split_sizes_0, x = var_2114_cast)[name = tensor<string, []>("op_2115_cast")];
            tensor<string, []> var_2117_mode_0 = const()[name = tensor<string, []>("op_2117_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_2117_cast = gelu(mode = var_2117_mode_0, x = var_2115_cast_1)[name = tensor<string, []>("op_2117_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_57_cast = mul(x = var_2115_cast_0, y = var_2117_cast)[name = tensor<string, []>("input_57_cast")];
            tensor<int32, [2]> var_2121 = const()[name = tensor<string, []>("op_2121"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2123 = const()[name = tensor<string, []>("op_2123"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2125_pad_type_0 = const()[name = tensor<string, []>("op_2125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2125_pad_0 = const()[name = tensor<string, []>("op_2125_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2902912))), lut = tensor<fp16, [4]>([-0x1.94cp-4, -0x1.e98p-6, 0x1.d6p-6, 0x1.8fcp-4]), 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, []>(3005376)))];
            tensor<fp16, [2, 320, 1, 4096]> var_2125_cast = conv(bias = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_2123, groups = var_120, pad = var_2125_pad_0, pad_type = var_2125_pad_type_0, strides = var_2121, weight = down_blocks_0_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_57_cast)[name = tensor<string, []>("op_2125_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_33_cast = add(x = var_2125_cast, y = inputs_11_cast)[name = tensor<string, []>("hidden_states_33_cast")];
            tensor<int32, [4]> var_2127 = const()[name = tensor<string, []>("op_2127"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_59_cast = reshape(shape = var_2127, x = hidden_states_33_cast)[name = tensor<string, []>("input_59_cast")];
            tensor<int32, [2]> var_2131 = const()[name = tensor<string, []>("op_2131"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2133 = const()[name = tensor<string, []>("op_2133"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3006080))), lut = tensor<fp16, [4]>([-0x1.b7p-4, -0x1.f14p-6, 0x1.134p-5, 0x1.c2cp-4]), 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, []>(3031744)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_35_cast = conv(bias = down_blocks_0_attentions_1_proj_out_bias_to_fp16, dilations = var_2133, groups = var_120, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = var_2131, 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_2140 = const()[name = tensor<string, []>("op_2140"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_2142 = const()[name = tensor<string, []>("op_2142"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3032448))), lut = tensor<fp16, [4]>([-0x1.b68p-5, -0x1.cfcp-7, 0x1.f78p-7, 0x1.c6cp-5]), 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, []>(3262912)))];
            tensor<fp16, [2, 320, 32, 32]> input_63_cast = conv(bias = down_blocks_0_downsamplers_0_conv_bias_to_fp16, dilations = var_2142, groups = var_120, pad = input_63_pad_0, pad_type = input_63_pad_type_0, strides = var_2140, weight = down_blocks_0_downsamplers_0_conv_weight_to_fp16_palettized, x = input_61_cast)[name = tensor<string, []>("input_63_cast")];
            tensor<int32, []> var_2150 = const()[name = tensor<string, []>("op_2150"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_2168 = const()[name = tensor<string, []>("op_2168"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_2173 = const()[name = tensor<string, []>("op_2173"), 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, []>(3263616)))];
            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, []>(3264320)))];
            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_2196 = const()[name = tensor<string, []>("op_2196"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2198 = const()[name = tensor<string, []>("op_2198"), 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, [460800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3265024))), lut = tensor<fp16, [4]>([-0x1.c58p-4, -0x1.eccp-6, 0x1.9cp-6, 0x1.ac4p-4]), 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, []>(3725888)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_37_cast = conv(bias = down_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_2198, groups = var_2173, pad = hidden_states_37_pad_0, pad_type = hidden_states_37_pad_type_0, strides = var_2196, 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_2204 = const()[name = tensor<string, []>("op_2204"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2206 = const()[name = tensor<string, []>("op_2206"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3727232))), lut = tensor<fp16, [4]>([-0x1.16p+1, -0x1.868p-2, -0x1.35p-8, 0x1.414p-8]), 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, []>(3932096)))];
            tensor<fp16, [2, 640, 1, 1]> temb_5_cast = conv(bias = down_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_2206, groups = var_2173, pad = temb_5_pad_0, pad_type = temb_5_pad_type_0, strides = var_2204, 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, []>(3933440)))];
            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, []>(3934784)))];
            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, []>(3936128)))];
            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, []>(3937472)))];
            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_2216 = const()[name = tensor<string, []>("op_2216"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2218 = const()[name = tensor<string, []>("op_2218"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3938816))), lut = tensor<fp16, [4]>([-0x1.644p-4, -0x1.728p-6, 0x1.774p-6, 0x1.65p-4]), 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, []>(4860480)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_39_cast = conv(bias = down_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_2218, groups = var_2173, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = var_2216, 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_2223 = const()[name = tensor<string, []>("op_2223"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2225 = const()[name = tensor<string, []>("op_2225"), 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, [51200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4861824))), lut = tensor<fp16, [4]>([-0x1.724p-5, -0x1.aacp-7, 0x1.c24p-7, 0x1.788p-5]), 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, []>(4913088)))];
            tensor<fp16, [2, 640, 32, 32]> x_1_cast = conv(bias = down_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_2225, groups = var_2173, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = var_2223, 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, []>(4914432)))];
            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, []>(4915776)))];
            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_2245 = const()[name = tensor<string, []>("op_2245"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2247 = const()[name = tensor<string, []>("op_2247"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4917120))), lut = tensor<fp16, [4]>([-0x1.94p-4, -0x1.e18p-6, 0x1.dbp-6, 0x1.91p-4]), 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, []>(5019584)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_43_cast = conv(bias = down_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_2247, groups = var_2173, pad = hidden_states_43_pad_0, pad_type = hidden_states_43_pad_type_0, strides = var_2245, 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_2252 = const()[name = tensor<string, []>("op_2252"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_13_cast = reshape(shape = var_2252, x = hidden_states_43_cast)[name = tensor<string, []>("inputs_13_cast")];
            tensor<int32, [1]> var_2262 = const()[name = tensor<string, []>("op_2262"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_13_cast = reduce_mean(axes = var_2262, keep_dims = var_2168, 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_2266 = const()[name = tensor<string, []>("op_2266"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_2267_cast = reduce_mean(axes = var_2266, keep_dims = var_2168, x = zero_mean_sq_13_cast)[name = tensor<string, []>("op_2267_cast")];
            tensor<fp16, []> var_2268_to_fp16 = const()[name = tensor<string, []>("op_2268_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_2269_cast = add(x = var_2267_cast, y = var_2268_to_fp16)[name = tensor<string, []>("op_2269_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_2269_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_2273_to_fp16 = const()[name = tensor<string, []>("op_2273_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5020928)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2274_cast = add(x = out_13_cast, y = var_2273_to_fp16)[name = tensor<string, []>("op_2274_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2276_to_fp16 = const()[name = tensor<string, []>("op_2276_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5022272)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_45_cast = mul(x = var_2274_cast, y = var_2276_to_fp16)[name = tensor<string, []>("hidden_states_45_cast")];
            tensor<int32, [2]> var_2283 = const()[name = tensor<string, []>("op_2283"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2285 = const()[name = tensor<string, []>("op_2285"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5023616))), lut = tensor<fp16, [4]>([-0x1.034p-3, -0x1.24cp-5, 0x1.34p-5, 0x1.07p-3]), 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_2285, groups = var_2173, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = var_2283, 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_2289 = const()[name = tensor<string, []>("op_2289"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2291 = const()[name = tensor<string, []>("op_2291"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5126080))), lut = tensor<fp16, [4]>([-0x1.f74p-4, -0x1.21p-5, 0x1.248p-5, 0x1.f9p-4]), 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_2291, groups = var_2173, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = var_2289, 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_2295 = const()[name = tensor<string, []>("op_2295"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2297 = const()[name = tensor<string, []>("op_2297"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5228544))), lut = tensor<fp16, [4]>([-0x1.68p-4, -0x1.a58p-6, 0x1.994p-6, 0x1.63cp-4]), 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_2297, groups = var_2173, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = var_2295, 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_2301_begin_0 = const()[name = tensor<string, []>("op_2301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2301_end_0 = const()[name = tensor<string, []>("op_2301_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2301_end_mask_0 = const()[name = tensor<string, []>("op_2301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2301_cast = slice_by_index(begin = var_2301_begin_0, end = var_2301_end_0, end_mask = var_2301_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2301_cast")];
            tensor<int32, [4]> var_2305_begin_0 = const()[name = tensor<string, []>("op_2305_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2305_end_0 = const()[name = tensor<string, []>("op_2305_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_2305_end_mask_0 = const()[name = tensor<string, []>("op_2305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2305_cast = slice_by_index(begin = var_2305_begin_0, end = var_2305_end_0, end_mask = var_2305_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2305_cast")];
            tensor<int32, [4]> var_2309_begin_0 = const()[name = tensor<string, []>("op_2309_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2309_end_0 = const()[name = tensor<string, []>("op_2309_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_2309_end_mask_0 = const()[name = tensor<string, []>("op_2309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2309_cast = slice_by_index(begin = var_2309_begin_0, end = var_2309_end_0, end_mask = var_2309_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2309_cast")];
            tensor<int32, [4]> var_2313_begin_0 = const()[name = tensor<string, []>("op_2313_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2313_end_0 = const()[name = tensor<string, []>("op_2313_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_2313_end_mask_0 = const()[name = tensor<string, []>("op_2313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2313_cast = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2313_cast")];
            tensor<int32, [4]> var_2317_begin_0 = const()[name = tensor<string, []>("op_2317_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2317_end_0 = const()[name = tensor<string, []>("op_2317_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_2317_end_mask_0 = const()[name = tensor<string, []>("op_2317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2317_cast = slice_by_index(begin = var_2317_begin_0, end = var_2317_end_0, end_mask = var_2317_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2317_cast")];
            tensor<int32, [4]> var_2321_begin_0 = const()[name = tensor<string, []>("op_2321_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2321_end_0 = const()[name = tensor<string, []>("op_2321_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_2321_end_mask_0 = const()[name = tensor<string, []>("op_2321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2321_cast = slice_by_index(begin = var_2321_begin_0, end = var_2321_end_0, end_mask = var_2321_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2321_cast")];
            tensor<int32, [4]> var_2325_begin_0 = const()[name = tensor<string, []>("op_2325_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2325_end_0 = const()[name = tensor<string, []>("op_2325_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_2325_end_mask_0 = const()[name = tensor<string, []>("op_2325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2325_cast = slice_by_index(begin = var_2325_begin_0, end = var_2325_end_0, end_mask = var_2325_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2325_cast")];
            tensor<int32, [4]> var_2329_begin_0 = const()[name = tensor<string, []>("op_2329_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2329_end_0 = const()[name = tensor<string, []>("op_2329_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_2329_end_mask_0 = const()[name = tensor<string, []>("op_2329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2329_cast = slice_by_index(begin = var_2329_begin_0, end = var_2329_end_0, end_mask = var_2329_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2329_cast")];
            tensor<int32, [4]> var_2333_begin_0 = const()[name = tensor<string, []>("op_2333_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2333_end_0 = const()[name = tensor<string, []>("op_2333_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_2333_end_mask_0 = const()[name = tensor<string, []>("op_2333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2333_cast = slice_by_index(begin = var_2333_begin_0, end = var_2333_end_0, end_mask = var_2333_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2333_cast")];
            tensor<int32, [4]> var_2337_begin_0 = const()[name = tensor<string, []>("op_2337_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2337_end_0 = const()[name = tensor<string, []>("op_2337_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_2337_end_mask_0 = const()[name = tensor<string, []>("op_2337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2337_cast = slice_by_index(begin = var_2337_begin_0, end = var_2337_end_0, end_mask = var_2337_end_mask_0, x = q_9_cast)[name = tensor<string, []>("op_2337_cast")];
            tensor<int32, [4]> var_2340_begin_0 = const()[name = tensor<string, []>("op_2340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2340_end_0 = const()[name = tensor<string, []>("op_2340_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2340_end_mask_0 = const()[name = tensor<string, []>("op_2340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2340_cast = slice_by_index(begin = var_2340_begin_0, end = var_2340_end_0, end_mask = var_2340_end_mask_0, x = var_2301_cast)[name = tensor<string, []>("op_2340_cast")];
            tensor<int32, [4]> var_2341_begin_0 = const()[name = tensor<string, []>("op_2341_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2341_end_0 = const()[name = tensor<string, []>("op_2341_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2341_end_mask_0 = const()[name = tensor<string, []>("op_2341_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2341_cast = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2301_cast)[name = tensor<string, []>("op_2341_cast")];
            tensor<int32, [4]> var_2342_begin_0 = const()[name = tensor<string, []>("op_2342_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2342_end_0 = const()[name = tensor<string, []>("op_2342_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2342_end_mask_0 = const()[name = tensor<string, []>("op_2342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2342_cast = slice_by_index(begin = var_2342_begin_0, end = var_2342_end_0, end_mask = var_2342_end_mask_0, x = var_2305_cast)[name = tensor<string, []>("op_2342_cast")];
            tensor<int32, [4]> var_2343_begin_0 = const()[name = tensor<string, []>("op_2343_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2343_end_0 = const()[name = tensor<string, []>("op_2343_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2343_end_mask_0 = const()[name = tensor<string, []>("op_2343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2343_cast = slice_by_index(begin = var_2343_begin_0, end = var_2343_end_0, end_mask = var_2343_end_mask_0, x = var_2305_cast)[name = tensor<string, []>("op_2343_cast")];
            tensor<int32, [4]> var_2344_begin_0 = const()[name = tensor<string, []>("op_2344_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2344_end_0 = const()[name = tensor<string, []>("op_2344_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2344_end_mask_0 = const()[name = tensor<string, []>("op_2344_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2344_cast = slice_by_index(begin = var_2344_begin_0, end = var_2344_end_0, end_mask = var_2344_end_mask_0, x = var_2309_cast)[name = tensor<string, []>("op_2344_cast")];
            tensor<int32, [4]> var_2345_begin_0 = const()[name = tensor<string, []>("op_2345_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2345_end_0 = const()[name = tensor<string, []>("op_2345_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2345_end_mask_0 = const()[name = tensor<string, []>("op_2345_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2345_cast = slice_by_index(begin = var_2345_begin_0, end = var_2345_end_0, end_mask = var_2345_end_mask_0, x = var_2309_cast)[name = tensor<string, []>("op_2345_cast")];
            tensor<int32, [4]> var_2346_begin_0 = const()[name = tensor<string, []>("op_2346_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2346_end_0 = const()[name = tensor<string, []>("op_2346_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2346_end_mask_0 = const()[name = tensor<string, []>("op_2346_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2346_cast = slice_by_index(begin = var_2346_begin_0, end = var_2346_end_0, end_mask = var_2346_end_mask_0, x = var_2313_cast)[name = tensor<string, []>("op_2346_cast")];
            tensor<int32, [4]> var_2347_begin_0 = const()[name = tensor<string, []>("op_2347_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2347_end_0 = const()[name = tensor<string, []>("op_2347_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2347_end_mask_0 = const()[name = tensor<string, []>("op_2347_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2347_cast = slice_by_index(begin = var_2347_begin_0, end = var_2347_end_0, end_mask = var_2347_end_mask_0, x = var_2313_cast)[name = tensor<string, []>("op_2347_cast")];
            tensor<int32, [4]> var_2348_begin_0 = const()[name = tensor<string, []>("op_2348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2348_end_0 = const()[name = tensor<string, []>("op_2348_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2348_end_mask_0 = const()[name = tensor<string, []>("op_2348_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2348_cast = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2317_cast)[name = tensor<string, []>("op_2348_cast")];
            tensor<int32, [4]> var_2349_begin_0 = const()[name = tensor<string, []>("op_2349_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2349_end_0 = const()[name = tensor<string, []>("op_2349_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2349_end_mask_0 = const()[name = tensor<string, []>("op_2349_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2349_cast = slice_by_index(begin = var_2349_begin_0, end = var_2349_end_0, end_mask = var_2349_end_mask_0, x = var_2317_cast)[name = tensor<string, []>("op_2349_cast")];
            tensor<int32, [4]> var_2350_begin_0 = const()[name = tensor<string, []>("op_2350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2350_end_0 = const()[name = tensor<string, []>("op_2350_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2350_end_mask_0 = const()[name = tensor<string, []>("op_2350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2350_cast = slice_by_index(begin = var_2350_begin_0, end = var_2350_end_0, end_mask = var_2350_end_mask_0, x = var_2321_cast)[name = tensor<string, []>("op_2350_cast")];
            tensor<int32, [4]> var_2351_begin_0 = const()[name = tensor<string, []>("op_2351_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2351_end_0 = const()[name = tensor<string, []>("op_2351_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2351_end_mask_0 = const()[name = tensor<string, []>("op_2351_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2351_cast = slice_by_index(begin = var_2351_begin_0, end = var_2351_end_0, end_mask = var_2351_end_mask_0, x = var_2321_cast)[name = tensor<string, []>("op_2351_cast")];
            tensor<int32, [4]> var_2352_begin_0 = const()[name = tensor<string, []>("op_2352_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2352_end_0 = const()[name = tensor<string, []>("op_2352_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2352_end_mask_0 = const()[name = tensor<string, []>("op_2352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2352_cast = slice_by_index(begin = var_2352_begin_0, end = var_2352_end_0, end_mask = var_2352_end_mask_0, x = var_2325_cast)[name = tensor<string, []>("op_2352_cast")];
            tensor<int32, [4]> var_2353_begin_0 = const()[name = tensor<string, []>("op_2353_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2353_end_0 = const()[name = tensor<string, []>("op_2353_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2353_end_mask_0 = const()[name = tensor<string, []>("op_2353_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2353_cast = slice_by_index(begin = var_2353_begin_0, end = var_2353_end_0, end_mask = var_2353_end_mask_0, x = var_2325_cast)[name = tensor<string, []>("op_2353_cast")];
            tensor<int32, [4]> var_2354_begin_0 = const()[name = tensor<string, []>("op_2354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2354_end_0 = const()[name = tensor<string, []>("op_2354_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2354_end_mask_0 = const()[name = tensor<string, []>("op_2354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2354_cast = slice_by_index(begin = var_2354_begin_0, end = var_2354_end_0, end_mask = var_2354_end_mask_0, x = var_2329_cast)[name = tensor<string, []>("op_2354_cast")];
            tensor<int32, [4]> var_2355_begin_0 = const()[name = tensor<string, []>("op_2355_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2355_end_0 = const()[name = tensor<string, []>("op_2355_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2355_end_mask_0 = const()[name = tensor<string, []>("op_2355_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2355_cast = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2329_cast)[name = tensor<string, []>("op_2355_cast")];
            tensor<int32, [4]> var_2356_begin_0 = const()[name = tensor<string, []>("op_2356_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2356_end_0 = const()[name = tensor<string, []>("op_2356_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2356_end_mask_0 = const()[name = tensor<string, []>("op_2356_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2356_cast = slice_by_index(begin = var_2356_begin_0, end = var_2356_end_0, end_mask = var_2356_end_mask_0, x = var_2333_cast)[name = tensor<string, []>("op_2356_cast")];
            tensor<int32, [4]> var_2357_begin_0 = const()[name = tensor<string, []>("op_2357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2357_end_0 = const()[name = tensor<string, []>("op_2357_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2357_end_mask_0 = const()[name = tensor<string, []>("op_2357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2357_cast = slice_by_index(begin = var_2357_begin_0, end = var_2357_end_0, end_mask = var_2357_end_mask_0, x = var_2333_cast)[name = tensor<string, []>("op_2357_cast")];
            tensor<int32, [4]> var_2358_begin_0 = const()[name = tensor<string, []>("op_2358_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2358_end_0 = const()[name = tensor<string, []>("op_2358_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2358_end_mask_0 = const()[name = tensor<string, []>("op_2358_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2358_cast = slice_by_index(begin = var_2358_begin_0, end = var_2358_end_0, end_mask = var_2358_end_mask_0, x = var_2337_cast)[name = tensor<string, []>("op_2358_cast")];
            tensor<int32, [4]> var_2359_begin_0 = const()[name = tensor<string, []>("op_2359_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2359_end_0 = const()[name = tensor<string, []>("op_2359_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2359_end_mask_0 = const()[name = tensor<string, []>("op_2359_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2359_cast = slice_by_index(begin = var_2359_begin_0, end = var_2359_end_0, end_mask = var_2359_end_mask_0, x = var_2337_cast)[name = tensor<string, []>("op_2359_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_2364_begin_0 = const()[name = tensor<string, []>("op_2364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2364_end_0 = const()[name = tensor<string, []>("op_2364_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_2364_end_mask_0 = const()[name = tensor<string, []>("op_2364_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, 64]> var_2364_cast = slice_by_index(begin = var_2364_begin_0, end = var_2364_end_0, end_mask = var_2364_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2364_cast")];
            tensor<int32, [4]> var_2368_begin_0 = const()[name = tensor<string, []>("op_2368_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2368_end_0 = const()[name = tensor<string, []>("op_2368_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_2368_end_mask_0 = const()[name = tensor<string, []>("op_2368_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2368_cast = slice_by_index(begin = var_2368_begin_0, end = var_2368_end_0, end_mask = var_2368_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2368_cast")];
            tensor<int32, [4]> var_2372_begin_0 = const()[name = tensor<string, []>("op_2372_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2372_end_0 = const()[name = tensor<string, []>("op_2372_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_2372_end_mask_0 = const()[name = tensor<string, []>("op_2372_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2372_cast = slice_by_index(begin = var_2372_begin_0, end = var_2372_end_0, end_mask = var_2372_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2372_cast")];
            tensor<int32, [4]> var_2376_begin_0 = const()[name = tensor<string, []>("op_2376_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2376_end_0 = const()[name = tensor<string, []>("op_2376_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_2376_end_mask_0 = const()[name = tensor<string, []>("op_2376_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2376_cast = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2376_cast")];
            tensor<int32, [4]> var_2380_begin_0 = const()[name = tensor<string, []>("op_2380_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2380_end_0 = const()[name = tensor<string, []>("op_2380_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_2380_end_mask_0 = const()[name = tensor<string, []>("op_2380_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2380_cast = slice_by_index(begin = var_2380_begin_0, end = var_2380_end_0, end_mask = var_2380_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2380_cast")];
            tensor<int32, [4]> var_2384_begin_0 = const()[name = tensor<string, []>("op_2384_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2384_end_0 = const()[name = tensor<string, []>("op_2384_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_2384_end_mask_0 = const()[name = tensor<string, []>("op_2384_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2384_cast = slice_by_index(begin = var_2384_begin_0, end = var_2384_end_0, end_mask = var_2384_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2384_cast")];
            tensor<int32, [4]> var_2388_begin_0 = const()[name = tensor<string, []>("op_2388_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2388_end_0 = const()[name = tensor<string, []>("op_2388_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_2388_end_mask_0 = const()[name = tensor<string, []>("op_2388_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2388_cast = slice_by_index(begin = var_2388_begin_0, end = var_2388_end_0, end_mask = var_2388_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2388_cast")];
            tensor<int32, [4]> var_2392_begin_0 = const()[name = tensor<string, []>("op_2392_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2392_end_0 = const()[name = tensor<string, []>("op_2392_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_2392_end_mask_0 = const()[name = tensor<string, []>("op_2392_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2392_cast = slice_by_index(begin = var_2392_begin_0, end = var_2392_end_0, end_mask = var_2392_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2392_cast")];
            tensor<int32, [4]> var_2396_begin_0 = const()[name = tensor<string, []>("op_2396_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2396_end_0 = const()[name = tensor<string, []>("op_2396_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_2396_end_mask_0 = const()[name = tensor<string, []>("op_2396_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2396_cast = slice_by_index(begin = var_2396_begin_0, end = var_2396_end_0, end_mask = var_2396_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2396_cast")];
            tensor<int32, [4]> var_2400_begin_0 = const()[name = tensor<string, []>("op_2400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2400_end_0 = const()[name = tensor<string, []>("op_2400_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_2400_end_mask_0 = const()[name = tensor<string, []>("op_2400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_2400_cast = slice_by_index(begin = var_2400_begin_0, end = var_2400_end_0, end_mask = var_2400_end_mask_0, x = transpose_27)[name = tensor<string, []>("op_2400_cast")];
            tensor<int32, [4]> var_2402_begin_0 = const()[name = tensor<string, []>("op_2402_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2402_end_0 = const()[name = tensor<string, []>("op_2402_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2402_end_mask_0 = const()[name = tensor<string, []>("op_2402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2402_cast = slice_by_index(begin = var_2402_begin_0, end = var_2402_end_0, end_mask = var_2402_end_mask_0, x = v_9_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, 64, 0, 0])];
            tensor<int32, [4]> var_2406_end_0 = const()[name = tensor<string, []>("op_2406_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_2406_end_mask_0 = const()[name = tensor<string, []>("op_2406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2406_cast = slice_by_index(begin = var_2406_begin_0, end = var_2406_end_0, end_mask = var_2406_end_mask_0, x = v_9_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, 128, 0, 0])];
            tensor<int32, [4]> var_2410_end_0 = const()[name = tensor<string, []>("op_2410_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_2410_end_mask_0 = const()[name = tensor<string, []>("op_2410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2410_cast = slice_by_index(begin = var_2410_begin_0, end = var_2410_end_0, end_mask = var_2410_end_mask_0, x = v_9_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, 192, 0, 0])];
            tensor<int32, [4]> var_2414_end_0 = const()[name = tensor<string, []>("op_2414_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_2414_end_mask_0 = const()[name = tensor<string, []>("op_2414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2414_cast = slice_by_index(begin = var_2414_begin_0, end = var_2414_end_0, end_mask = var_2414_end_mask_0, x = v_9_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, 256, 0, 0])];
            tensor<int32, [4]> var_2418_end_0 = const()[name = tensor<string, []>("op_2418_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_2418_end_mask_0 = const()[name = tensor<string, []>("op_2418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2418_cast = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_2418_cast")];
            tensor<int32, [4]> var_2422_begin_0 = const()[name = tensor<string, []>("op_2422_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2422_end_0 = const()[name = tensor<string, []>("op_2422_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_2422_end_mask_0 = const()[name = tensor<string, []>("op_2422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2422_cast = slice_by_index(begin = var_2422_begin_0, end = var_2422_end_0, end_mask = var_2422_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_2422_cast")];
            tensor<int32, [4]> var_2426_begin_0 = const()[name = tensor<string, []>("op_2426_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2426_end_0 = const()[name = tensor<string, []>("op_2426_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_2426_end_mask_0 = const()[name = tensor<string, []>("op_2426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2426_cast = slice_by_index(begin = var_2426_begin_0, end = var_2426_end_0, end_mask = var_2426_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_2426_cast")];
            tensor<int32, [4]> var_2430_begin_0 = const()[name = tensor<string, []>("op_2430_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2430_end_0 = const()[name = tensor<string, []>("op_2430_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_2430_end_mask_0 = const()[name = tensor<string, []>("op_2430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2430_cast = slice_by_index(begin = var_2430_begin_0, end = var_2430_end_0, end_mask = var_2430_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_2430_cast")];
            tensor<int32, [4]> var_2434_begin_0 = const()[name = tensor<string, []>("op_2434_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2434_end_0 = const()[name = tensor<string, []>("op_2434_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_2434_end_mask_0 = const()[name = tensor<string, []>("op_2434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2434_cast = slice_by_index(begin = var_2434_begin_0, end = var_2434_end_0, end_mask = var_2434_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_2434_cast")];
            tensor<int32, [4]> var_2438_begin_0 = const()[name = tensor<string, []>("op_2438_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2438_end_0 = const()[name = tensor<string, []>("op_2438_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_2438_end_mask_0 = const()[name = tensor<string, []>("op_2438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2438_cast = slice_by_index(begin = var_2438_begin_0, end = var_2438_end_0, end_mask = var_2438_end_mask_0, x = v_9_cast)[name = tensor<string, []>("op_2438_cast")];
            tensor<string, []> var_2442_equation_0 = const()[name = tensor<string, []>("op_2442_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2442_cast = einsum(equation = var_2442_equation_0, values = (var_2364_cast, var_2340_cast))[name = tensor<string, []>("op_2442_cast")];
            tensor<fp16, []> var_2443_to_fp16 = const()[name = tensor<string, []>("op_2443_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_321_cast = mul(x = var_2442_cast, y = var_2443_to_fp16)[name = tensor<string, []>("aw_chunk_321_cast")];
            tensor<string, []> var_2446_equation_0 = const()[name = tensor<string, []>("op_2446_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2446_cast = einsum(equation = var_2446_equation_0, values = (var_2364_cast, var_2341_cast))[name = tensor<string, []>("op_2446_cast")];
            tensor<fp16, []> var_2447_to_fp16 = const()[name = tensor<string, []>("op_2447_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_323_cast = mul(x = var_2446_cast, y = var_2447_to_fp16)[name = tensor<string, []>("aw_chunk_323_cast")];
            tensor<string, []> var_2450_equation_0 = const()[name = tensor<string, []>("op_2450_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2450_cast = einsum(equation = var_2450_equation_0, values = (var_2368_cast, var_2342_cast))[name = tensor<string, []>("op_2450_cast")];
            tensor<fp16, []> var_2451_to_fp16 = const()[name = tensor<string, []>("op_2451_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_325_cast = mul(x = var_2450_cast, y = var_2451_to_fp16)[name = tensor<string, []>("aw_chunk_325_cast")];
            tensor<string, []> var_2454_equation_0 = const()[name = tensor<string, []>("op_2454_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2454_cast = einsum(equation = var_2454_equation_0, values = (var_2368_cast, var_2343_cast))[name = tensor<string, []>("op_2454_cast")];
            tensor<fp16, []> var_2455_to_fp16 = const()[name = tensor<string, []>("op_2455_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_327_cast = mul(x = var_2454_cast, y = var_2455_to_fp16)[name = tensor<string, []>("aw_chunk_327_cast")];
            tensor<string, []> var_2458_equation_0 = const()[name = tensor<string, []>("op_2458_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2458_cast = einsum(equation = var_2458_equation_0, values = (var_2372_cast, var_2344_cast))[name = tensor<string, []>("op_2458_cast")];
            tensor<fp16, []> var_2459_to_fp16 = const()[name = tensor<string, []>("op_2459_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_329_cast = mul(x = var_2458_cast, y = var_2459_to_fp16)[name = tensor<string, []>("aw_chunk_329_cast")];
            tensor<string, []> var_2462_equation_0 = const()[name = tensor<string, []>("op_2462_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2462_cast = einsum(equation = var_2462_equation_0, values = (var_2372_cast, var_2345_cast))[name = tensor<string, []>("op_2462_cast")];
            tensor<fp16, []> var_2463_to_fp16 = const()[name = tensor<string, []>("op_2463_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_331_cast = mul(x = var_2462_cast, y = var_2463_to_fp16)[name = tensor<string, []>("aw_chunk_331_cast")];
            tensor<string, []> var_2466_equation_0 = const()[name = tensor<string, []>("op_2466_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2466_cast = einsum(equation = var_2466_equation_0, values = (var_2376_cast, var_2346_cast))[name = tensor<string, []>("op_2466_cast")];
            tensor<fp16, []> var_2467_to_fp16 = const()[name = tensor<string, []>("op_2467_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_333_cast = mul(x = var_2466_cast, y = var_2467_to_fp16)[name = tensor<string, []>("aw_chunk_333_cast")];
            tensor<string, []> var_2470_equation_0 = const()[name = tensor<string, []>("op_2470_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2470_cast = einsum(equation = var_2470_equation_0, values = (var_2376_cast, var_2347_cast))[name = tensor<string, []>("op_2470_cast")];
            tensor<fp16, []> var_2471_to_fp16 = const()[name = tensor<string, []>("op_2471_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_335_cast = mul(x = var_2470_cast, y = var_2471_to_fp16)[name = tensor<string, []>("aw_chunk_335_cast")];
            tensor<string, []> var_2474_equation_0 = const()[name = tensor<string, []>("op_2474_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2474_cast = einsum(equation = var_2474_equation_0, values = (var_2380_cast, var_2348_cast))[name = tensor<string, []>("op_2474_cast")];
            tensor<fp16, []> var_2475_to_fp16 = const()[name = tensor<string, []>("op_2475_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_337_cast = mul(x = var_2474_cast, y = var_2475_to_fp16)[name = tensor<string, []>("aw_chunk_337_cast")];
            tensor<string, []> var_2478_equation_0 = const()[name = tensor<string, []>("op_2478_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2478_cast = einsum(equation = var_2478_equation_0, values = (var_2380_cast, var_2349_cast))[name = tensor<string, []>("op_2478_cast")];
            tensor<fp16, []> var_2479_to_fp16 = const()[name = tensor<string, []>("op_2479_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_339_cast = mul(x = var_2478_cast, y = var_2479_to_fp16)[name = tensor<string, []>("aw_chunk_339_cast")];
            tensor<string, []> var_2482_equation_0 = const()[name = tensor<string, []>("op_2482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2482_cast = einsum(equation = var_2482_equation_0, values = (var_2384_cast, var_2350_cast))[name = tensor<string, []>("op_2482_cast")];
            tensor<fp16, []> var_2483_to_fp16 = const()[name = tensor<string, []>("op_2483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_341_cast = mul(x = var_2482_cast, y = var_2483_to_fp16)[name = tensor<string, []>("aw_chunk_341_cast")];
            tensor<string, []> var_2486_equation_0 = const()[name = tensor<string, []>("op_2486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2486_cast = einsum(equation = var_2486_equation_0, values = (var_2384_cast, var_2351_cast))[name = tensor<string, []>("op_2486_cast")];
            tensor<fp16, []> var_2487_to_fp16 = const()[name = tensor<string, []>("op_2487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_343_cast = mul(x = var_2486_cast, y = var_2487_to_fp16)[name = tensor<string, []>("aw_chunk_343_cast")];
            tensor<string, []> var_2490_equation_0 = const()[name = tensor<string, []>("op_2490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2490_cast = einsum(equation = var_2490_equation_0, values = (var_2388_cast, var_2352_cast))[name = tensor<string, []>("op_2490_cast")];
            tensor<fp16, []> var_2491_to_fp16 = const()[name = tensor<string, []>("op_2491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_345_cast = mul(x = var_2490_cast, y = var_2491_to_fp16)[name = tensor<string, []>("aw_chunk_345_cast")];
            tensor<string, []> var_2494_equation_0 = const()[name = tensor<string, []>("op_2494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2494_cast = einsum(equation = var_2494_equation_0, values = (var_2388_cast, var_2353_cast))[name = tensor<string, []>("op_2494_cast")];
            tensor<fp16, []> var_2495_to_fp16 = const()[name = tensor<string, []>("op_2495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_347_cast = mul(x = var_2494_cast, y = var_2495_to_fp16)[name = tensor<string, []>("aw_chunk_347_cast")];
            tensor<string, []> var_2498_equation_0 = const()[name = tensor<string, []>("op_2498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2498_cast = einsum(equation = var_2498_equation_0, values = (var_2392_cast, var_2354_cast))[name = tensor<string, []>("op_2498_cast")];
            tensor<fp16, []> var_2499_to_fp16 = const()[name = tensor<string, []>("op_2499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_349_cast = mul(x = var_2498_cast, y = var_2499_to_fp16)[name = tensor<string, []>("aw_chunk_349_cast")];
            tensor<string, []> var_2502_equation_0 = const()[name = tensor<string, []>("op_2502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2502_cast = einsum(equation = var_2502_equation_0, values = (var_2392_cast, var_2355_cast))[name = tensor<string, []>("op_2502_cast")];
            tensor<fp16, []> var_2503_to_fp16 = const()[name = tensor<string, []>("op_2503_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_351_cast = mul(x = var_2502_cast, y = var_2503_to_fp16)[name = tensor<string, []>("aw_chunk_351_cast")];
            tensor<string, []> var_2506_equation_0 = const()[name = tensor<string, []>("op_2506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2506_cast = einsum(equation = var_2506_equation_0, values = (var_2396_cast, var_2356_cast))[name = tensor<string, []>("op_2506_cast")];
            tensor<fp16, []> var_2507_to_fp16 = const()[name = tensor<string, []>("op_2507_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_353_cast = mul(x = var_2506_cast, y = var_2507_to_fp16)[name = tensor<string, []>("aw_chunk_353_cast")];
            tensor<string, []> var_2510_equation_0 = const()[name = tensor<string, []>("op_2510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2510_cast = einsum(equation = var_2510_equation_0, values = (var_2396_cast, var_2357_cast))[name = tensor<string, []>("op_2510_cast")];
            tensor<fp16, []> var_2511_to_fp16 = const()[name = tensor<string, []>("op_2511_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_355_cast = mul(x = var_2510_cast, y = var_2511_to_fp16)[name = tensor<string, []>("aw_chunk_355_cast")];
            tensor<string, []> var_2514_equation_0 = const()[name = tensor<string, []>("op_2514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2514_cast = einsum(equation = var_2514_equation_0, values = (var_2400_cast, var_2358_cast))[name = tensor<string, []>("op_2514_cast")];
            tensor<fp16, []> var_2515_to_fp16 = const()[name = tensor<string, []>("op_2515_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_357_cast = mul(x = var_2514_cast, y = var_2515_to_fp16)[name = tensor<string, []>("aw_chunk_357_cast")];
            tensor<string, []> var_2518_equation_0 = const()[name = tensor<string, []>("op_2518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_2518_cast = einsum(equation = var_2518_equation_0, values = (var_2400_cast, var_2359_cast))[name = tensor<string, []>("op_2518_cast")];
            tensor<fp16, []> var_2519_to_fp16 = const()[name = tensor<string, []>("op_2519_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_359_cast = mul(x = var_2518_cast, y = var_2519_to_fp16)[name = tensor<string, []>("aw_chunk_359_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2521_cast = softmax(axis = var_2173, x = aw_chunk_321_cast)[name = tensor<string, []>("op_2521_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2522_cast = softmax(axis = var_2173, x = aw_chunk_323_cast)[name = tensor<string, []>("op_2522_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2523_cast = softmax(axis = var_2173, x = aw_chunk_325_cast)[name = tensor<string, []>("op_2523_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2524_cast = softmax(axis = var_2173, x = aw_chunk_327_cast)[name = tensor<string, []>("op_2524_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2525_cast = softmax(axis = var_2173, x = aw_chunk_329_cast)[name = tensor<string, []>("op_2525_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2526_cast = softmax(axis = var_2173, x = aw_chunk_331_cast)[name = tensor<string, []>("op_2526_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2527_cast = softmax(axis = var_2173, x = aw_chunk_333_cast)[name = tensor<string, []>("op_2527_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2528_cast = softmax(axis = var_2173, x = aw_chunk_335_cast)[name = tensor<string, []>("op_2528_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2529_cast = softmax(axis = var_2173, x = aw_chunk_337_cast)[name = tensor<string, []>("op_2529_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2530_cast = softmax(axis = var_2173, x = aw_chunk_339_cast)[name = tensor<string, []>("op_2530_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2531_cast = softmax(axis = var_2173, x = aw_chunk_341_cast)[name = tensor<string, []>("op_2531_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2532_cast = softmax(axis = var_2173, x = aw_chunk_343_cast)[name = tensor<string, []>("op_2532_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2533_cast = softmax(axis = var_2173, x = aw_chunk_345_cast)[name = tensor<string, []>("op_2533_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2534_cast = softmax(axis = var_2173, x = aw_chunk_347_cast)[name = tensor<string, []>("op_2534_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2535_cast = softmax(axis = var_2173, x = aw_chunk_349_cast)[name = tensor<string, []>("op_2535_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2536_cast = softmax(axis = var_2173, x = aw_chunk_351_cast)[name = tensor<string, []>("op_2536_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2537_cast = softmax(axis = var_2173, x = aw_chunk_353_cast)[name = tensor<string, []>("op_2537_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2538_cast = softmax(axis = var_2173, x = aw_chunk_355_cast)[name = tensor<string, []>("op_2538_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2539_cast = softmax(axis = var_2173, x = aw_chunk_357_cast)[name = tensor<string, []>("op_2539_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_2540_cast = softmax(axis = var_2173, x = aw_chunk_359_cast)[name = tensor<string, []>("op_2540_cast")];
            tensor<string, []> var_2542_equation_0 = const()[name = tensor<string, []>("op_2542_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2542_cast = einsum(equation = var_2542_equation_0, values = (var_2402_cast, var_2521_cast))[name = tensor<string, []>("op_2542_cast")];
            tensor<string, []> var_2544_equation_0 = const()[name = tensor<string, []>("op_2544_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2544_cast = einsum(equation = var_2544_equation_0, values = (var_2402_cast, var_2522_cast))[name = tensor<string, []>("op_2544_cast")];
            tensor<string, []> var_2546_equation_0 = const()[name = tensor<string, []>("op_2546_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2546_cast = einsum(equation = var_2546_equation_0, values = (var_2406_cast, var_2523_cast))[name = tensor<string, []>("op_2546_cast")];
            tensor<string, []> var_2548_equation_0 = const()[name = tensor<string, []>("op_2548_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2548_cast = einsum(equation = var_2548_equation_0, values = (var_2406_cast, var_2524_cast))[name = tensor<string, []>("op_2548_cast")];
            tensor<string, []> var_2550_equation_0 = const()[name = tensor<string, []>("op_2550_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2550_cast = einsum(equation = var_2550_equation_0, values = (var_2410_cast, var_2525_cast))[name = tensor<string, []>("op_2550_cast")];
            tensor<string, []> var_2552_equation_0 = const()[name = tensor<string, []>("op_2552_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2552_cast = einsum(equation = var_2552_equation_0, values = (var_2410_cast, var_2526_cast))[name = tensor<string, []>("op_2552_cast")];
            tensor<string, []> var_2554_equation_0 = const()[name = tensor<string, []>("op_2554_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2554_cast = einsum(equation = var_2554_equation_0, values = (var_2414_cast, var_2527_cast))[name = tensor<string, []>("op_2554_cast")];
            tensor<string, []> var_2556_equation_0 = const()[name = tensor<string, []>("op_2556_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2556_cast = einsum(equation = var_2556_equation_0, values = (var_2414_cast, var_2528_cast))[name = tensor<string, []>("op_2556_cast")];
            tensor<string, []> var_2558_equation_0 = const()[name = tensor<string, []>("op_2558_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2558_cast = einsum(equation = var_2558_equation_0, values = (var_2418_cast, var_2529_cast))[name = tensor<string, []>("op_2558_cast")];
            tensor<string, []> var_2560_equation_0 = const()[name = tensor<string, []>("op_2560_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2560_cast = einsum(equation = var_2560_equation_0, values = (var_2418_cast, var_2530_cast))[name = tensor<string, []>("op_2560_cast")];
            tensor<string, []> var_2562_equation_0 = const()[name = tensor<string, []>("op_2562_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2562_cast = einsum(equation = var_2562_equation_0, values = (var_2422_cast, var_2531_cast))[name = tensor<string, []>("op_2562_cast")];
            tensor<string, []> var_2564_equation_0 = const()[name = tensor<string, []>("op_2564_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2564_cast = einsum(equation = var_2564_equation_0, values = (var_2422_cast, var_2532_cast))[name = tensor<string, []>("op_2564_cast")];
            tensor<string, []> var_2566_equation_0 = const()[name = tensor<string, []>("op_2566_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2566_cast = einsum(equation = var_2566_equation_0, values = (var_2426_cast, var_2533_cast))[name = tensor<string, []>("op_2566_cast")];
            tensor<string, []> var_2568_equation_0 = const()[name = tensor<string, []>("op_2568_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2568_cast = einsum(equation = var_2568_equation_0, values = (var_2426_cast, var_2534_cast))[name = tensor<string, []>("op_2568_cast")];
            tensor<string, []> var_2570_equation_0 = const()[name = tensor<string, []>("op_2570_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2570_cast = einsum(equation = var_2570_equation_0, values = (var_2430_cast, var_2535_cast))[name = tensor<string, []>("op_2570_cast")];
            tensor<string, []> var_2572_equation_0 = const()[name = tensor<string, []>("op_2572_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2572_cast = einsum(equation = var_2572_equation_0, values = (var_2430_cast, var_2536_cast))[name = tensor<string, []>("op_2572_cast")];
            tensor<string, []> var_2574_equation_0 = const()[name = tensor<string, []>("op_2574_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2574_cast = einsum(equation = var_2574_equation_0, values = (var_2434_cast, var_2537_cast))[name = tensor<string, []>("op_2574_cast")];
            tensor<string, []> var_2576_equation_0 = const()[name = tensor<string, []>("op_2576_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2576_cast = einsum(equation = var_2576_equation_0, values = (var_2434_cast, var_2538_cast))[name = tensor<string, []>("op_2576_cast")];
            tensor<string, []> var_2578_equation_0 = const()[name = tensor<string, []>("op_2578_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2578_cast = einsum(equation = var_2578_equation_0, values = (var_2438_cast, var_2539_cast))[name = tensor<string, []>("op_2578_cast")];
            tensor<string, []> var_2580_equation_0 = const()[name = tensor<string, []>("op_2580_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2580_cast = einsum(equation = var_2580_equation_0, values = (var_2438_cast, var_2540_cast))[name = tensor<string, []>("op_2580_cast")];
            tensor<bool, []> var_2582_interleave_0 = const()[name = tensor<string, []>("op_2582_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2582_cast = concat(axis = var_2150, interleave = var_2582_interleave_0, values = (var_2542_cast, var_2544_cast))[name = tensor<string, []>("op_2582_cast")];
            tensor<bool, []> var_2584_interleave_0 = const()[name = tensor<string, []>("op_2584_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2584_cast = concat(axis = var_2150, interleave = var_2584_interleave_0, values = (var_2546_cast, var_2548_cast))[name = tensor<string, []>("op_2584_cast")];
            tensor<bool, []> var_2586_interleave_0 = const()[name = tensor<string, []>("op_2586_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2586_cast = concat(axis = var_2150, interleave = var_2586_interleave_0, values = (var_2550_cast, var_2552_cast))[name = tensor<string, []>("op_2586_cast")];
            tensor<bool, []> var_2588_interleave_0 = const()[name = tensor<string, []>("op_2588_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2588_cast = concat(axis = var_2150, interleave = var_2588_interleave_0, values = (var_2554_cast, var_2556_cast))[name = tensor<string, []>("op_2588_cast")];
            tensor<bool, []> var_2590_interleave_0 = const()[name = tensor<string, []>("op_2590_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2590_cast = concat(axis = var_2150, interleave = var_2590_interleave_0, values = (var_2558_cast, var_2560_cast))[name = tensor<string, []>("op_2590_cast")];
            tensor<bool, []> var_2592_interleave_0 = const()[name = tensor<string, []>("op_2592_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2592_cast = concat(axis = var_2150, interleave = var_2592_interleave_0, values = (var_2562_cast, var_2564_cast))[name = tensor<string, []>("op_2592_cast")];
            tensor<bool, []> var_2594_interleave_0 = const()[name = tensor<string, []>("op_2594_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2594_cast = concat(axis = var_2150, interleave = var_2594_interleave_0, values = (var_2566_cast, var_2568_cast))[name = tensor<string, []>("op_2594_cast")];
            tensor<bool, []> var_2596_interleave_0 = const()[name = tensor<string, []>("op_2596_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2596_cast = concat(axis = var_2150, interleave = var_2596_interleave_0, values = (var_2570_cast, var_2572_cast))[name = tensor<string, []>("op_2596_cast")];
            tensor<bool, []> var_2598_interleave_0 = const()[name = tensor<string, []>("op_2598_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2598_cast = concat(axis = var_2150, interleave = var_2598_interleave_0, values = (var_2574_cast, var_2576_cast))[name = tensor<string, []>("op_2598_cast")];
            tensor<bool, []> var_2600_interleave_0 = const()[name = tensor<string, []>("op_2600_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2600_cast = concat(axis = var_2150, interleave = var_2600_interleave_0, values = (var_2578_cast, var_2580_cast))[name = tensor<string, []>("op_2600_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_2173, interleave = input_79_interleave_0, values = (var_2582_cast, var_2584_cast, var_2586_cast, var_2588_cast, var_2590_cast, var_2592_cast, var_2594_cast, var_2596_cast, var_2598_cast, var_2600_cast))[name = tensor<string, []>("input_79_cast")];
            tensor<int32, [2]> var_2606 = const()[name = tensor<string, []>("op_2606"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2608 = const()[name = tensor<string, []>("op_2608"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2610_pad_type_0 = const()[name = tensor<string, []>("op_2610_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2610_pad_0 = const()[name = tensor<string, []>("op_2610_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5331008))), lut = tensor<fp16, [4]>([-0x1.5d8p-4, -0x1.a2p-6, 0x1.9b4p-6, 0x1.5bp-4]), 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, []>(5433472)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2610_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_2608, groups = var_2173, pad = var_2610_pad_0, pad_type = var_2610_pad_type_0, strides = var_2606, 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_2610_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_15_cast = add(x = var_2610_cast, y = inputs_13_cast)[name = tensor<string, []>("inputs_15_cast")];
            tensor<int32, [1]> var_2614 = const()[name = tensor<string, []>("op_2614"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_15_cast = reduce_mean(axes = var_2614, keep_dims = var_2168, 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_2618 = const()[name = tensor<string, []>("op_2618"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_2619_cast = reduce_mean(axes = var_2618, keep_dims = var_2168, x = zero_mean_sq_15_cast)[name = tensor<string, []>("op_2619_cast")];
            tensor<fp16, []> var_2620_to_fp16 = const()[name = tensor<string, []>("op_2620_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_2621_cast = add(x = var_2619_cast, y = var_2620_to_fp16)[name = tensor<string, []>("op_2621_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_2621_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_2625_to_fp16 = const()[name = tensor<string, []>("op_2625_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5434816)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2626_cast = add(x = out_15_cast, y = var_2625_to_fp16)[name = tensor<string, []>("op_2626_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2628_to_fp16 = const()[name = tensor<string, []>("op_2628_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5436160)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_47_cast = mul(x = var_2626_cast, y = var_2628_to_fp16)[name = tensor<string, []>("hidden_states_47_cast")];
            tensor<int32, [2]> var_2635 = const()[name = tensor<string, []>("op_2635"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2637 = const()[name = tensor<string, []>("op_2637"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5437504))), lut = tensor<fp16, [4]>([-0x1.844p-4, -0x1.ca4p-6, 0x1.d24p-6, 0x1.858p-4]), 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_2637, groups = var_2173, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = var_2635, 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_2641 = const()[name = tensor<string, []>("op_2641"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2643 = const()[name = tensor<string, []>("op_2643"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_21_pad_type_0 = const()[name = tensor<string, []>("k_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_21_pad_0 = const()[name = tensor<string, []>("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5539968))), lut = tensor<fp16, [4]>([-0x1.668p-4, -0x1.a28p-6, 0x1.adcp-6, 0x1.694p-4]), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_21_cast = conv(dilations = var_2643, groups = var_2173, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = var_2641, 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_2647 = const()[name = tensor<string, []>("op_2647"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2649 = const()[name = tensor<string, []>("op_2649"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_11_pad_type_0 = const()[name = tensor<string, []>("v_11_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_11_pad_0 = const()[name = tensor<string, []>("v_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5703872))), lut = tensor<fp16, [4]>([-0x1.394p-5, -0x1.72cp-7, 0x1.6b4p-7, 0x1.368p-5]), name = tensor<string, []>("down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_11_cast = conv(dilations = var_2649, groups = var_2173, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = var_2647, 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_2653_begin_0 = const()[name = tensor<string, []>("op_2653_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2653_end_0 = const()[name = tensor<string, []>("op_2653_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2653_end_mask_0 = const()[name = tensor<string, []>("op_2653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2653_cast = slice_by_index(begin = var_2653_begin_0, end = var_2653_end_0, end_mask = var_2653_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2653_cast")];
            tensor<int32, [4]> var_2657_begin_0 = const()[name = tensor<string, []>("op_2657_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2657_end_0 = const()[name = tensor<string, []>("op_2657_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_2657_end_mask_0 = const()[name = tensor<string, []>("op_2657_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2657_cast = slice_by_index(begin = var_2657_begin_0, end = var_2657_end_0, end_mask = var_2657_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2657_cast")];
            tensor<int32, [4]> var_2661_begin_0 = const()[name = tensor<string, []>("op_2661_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2661_end_0 = const()[name = tensor<string, []>("op_2661_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_2661_end_mask_0 = const()[name = tensor<string, []>("op_2661_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2661_cast = slice_by_index(begin = var_2661_begin_0, end = var_2661_end_0, end_mask = var_2661_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2661_cast")];
            tensor<int32, [4]> var_2665_begin_0 = const()[name = tensor<string, []>("op_2665_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2665_end_0 = const()[name = tensor<string, []>("op_2665_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_2665_end_mask_0 = const()[name = tensor<string, []>("op_2665_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2665_cast = slice_by_index(begin = var_2665_begin_0, end = var_2665_end_0, end_mask = var_2665_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2665_cast")];
            tensor<int32, [4]> var_2669_begin_0 = const()[name = tensor<string, []>("op_2669_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2669_end_0 = const()[name = tensor<string, []>("op_2669_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_2669_end_mask_0 = const()[name = tensor<string, []>("op_2669_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2669_cast = slice_by_index(begin = var_2669_begin_0, end = var_2669_end_0, end_mask = var_2669_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2669_cast")];
            tensor<int32, [4]> var_2673_begin_0 = const()[name = tensor<string, []>("op_2673_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2673_end_0 = const()[name = tensor<string, []>("op_2673_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_2673_end_mask_0 = const()[name = tensor<string, []>("op_2673_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2673_cast = slice_by_index(begin = var_2673_begin_0, end = var_2673_end_0, end_mask = var_2673_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2673_cast")];
            tensor<int32, [4]> var_2677_begin_0 = const()[name = tensor<string, []>("op_2677_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2677_end_0 = const()[name = tensor<string, []>("op_2677_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_2677_end_mask_0 = const()[name = tensor<string, []>("op_2677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2677_cast = slice_by_index(begin = var_2677_begin_0, end = var_2677_end_0, end_mask = var_2677_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2677_cast")];
            tensor<int32, [4]> var_2681_begin_0 = const()[name = tensor<string, []>("op_2681_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2681_end_0 = const()[name = tensor<string, []>("op_2681_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_2681_end_mask_0 = const()[name = tensor<string, []>("op_2681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2681_cast = slice_by_index(begin = var_2681_begin_0, end = var_2681_end_0, end_mask = var_2681_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2681_cast")];
            tensor<int32, [4]> var_2685_begin_0 = const()[name = tensor<string, []>("op_2685_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2685_end_0 = const()[name = tensor<string, []>("op_2685_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_2685_end_mask_0 = const()[name = tensor<string, []>("op_2685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2685_cast = slice_by_index(begin = var_2685_begin_0, end = var_2685_end_0, end_mask = var_2685_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2685_cast")];
            tensor<int32, [4]> var_2689_begin_0 = const()[name = tensor<string, []>("op_2689_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2689_end_0 = const()[name = tensor<string, []>("op_2689_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_2689_end_mask_0 = const()[name = tensor<string, []>("op_2689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_2689_cast = slice_by_index(begin = var_2689_begin_0, end = var_2689_end_0, end_mask = var_2689_end_mask_0, x = q_11_cast)[name = tensor<string, []>("op_2689_cast")];
            tensor<int32, [4]> var_2692_begin_0 = const()[name = tensor<string, []>("op_2692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2692_end_0 = const()[name = tensor<string, []>("op_2692_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2692_end_mask_0 = const()[name = tensor<string, []>("op_2692_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2692_cast = slice_by_index(begin = var_2692_begin_0, end = var_2692_end_0, end_mask = var_2692_end_mask_0, x = var_2653_cast)[name = tensor<string, []>("op_2692_cast")];
            tensor<int32, [4]> var_2693_begin_0 = const()[name = tensor<string, []>("op_2693_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2693_end_0 = const()[name = tensor<string, []>("op_2693_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2693_end_mask_0 = const()[name = tensor<string, []>("op_2693_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2693_cast = slice_by_index(begin = var_2693_begin_0, end = var_2693_end_0, end_mask = var_2693_end_mask_0, x = var_2653_cast)[name = tensor<string, []>("op_2693_cast")];
            tensor<int32, [4]> var_2694_begin_0 = const()[name = tensor<string, []>("op_2694_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2694_end_0 = const()[name = tensor<string, []>("op_2694_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2694_end_mask_0 = const()[name = tensor<string, []>("op_2694_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2694_cast = slice_by_index(begin = var_2694_begin_0, end = var_2694_end_0, end_mask = var_2694_end_mask_0, x = var_2657_cast)[name = tensor<string, []>("op_2694_cast")];
            tensor<int32, [4]> var_2695_begin_0 = const()[name = tensor<string, []>("op_2695_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2695_end_0 = const()[name = tensor<string, []>("op_2695_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2695_end_mask_0 = const()[name = tensor<string, []>("op_2695_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2695_cast = slice_by_index(begin = var_2695_begin_0, end = var_2695_end_0, end_mask = var_2695_end_mask_0, x = var_2657_cast)[name = tensor<string, []>("op_2695_cast")];
            tensor<int32, [4]> var_2696_begin_0 = const()[name = tensor<string, []>("op_2696_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2696_end_0 = const()[name = tensor<string, []>("op_2696_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2696_end_mask_0 = const()[name = tensor<string, []>("op_2696_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2696_cast = slice_by_index(begin = var_2696_begin_0, end = var_2696_end_0, end_mask = var_2696_end_mask_0, x = var_2661_cast)[name = tensor<string, []>("op_2696_cast")];
            tensor<int32, [4]> var_2697_begin_0 = const()[name = tensor<string, []>("op_2697_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2697_end_0 = const()[name = tensor<string, []>("op_2697_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2697_end_mask_0 = const()[name = tensor<string, []>("op_2697_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2697_cast = slice_by_index(begin = var_2697_begin_0, end = var_2697_end_0, end_mask = var_2697_end_mask_0, x = var_2661_cast)[name = tensor<string, []>("op_2697_cast")];
            tensor<int32, [4]> var_2698_begin_0 = const()[name = tensor<string, []>("op_2698_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2698_end_0 = const()[name = tensor<string, []>("op_2698_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2698_end_mask_0 = const()[name = tensor<string, []>("op_2698_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2698_cast = slice_by_index(begin = var_2698_begin_0, end = var_2698_end_0, end_mask = var_2698_end_mask_0, x = var_2665_cast)[name = tensor<string, []>("op_2698_cast")];
            tensor<int32, [4]> var_2699_begin_0 = const()[name = tensor<string, []>("op_2699_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2699_end_0 = const()[name = tensor<string, []>("op_2699_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2699_end_mask_0 = const()[name = tensor<string, []>("op_2699_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2699_cast = slice_by_index(begin = var_2699_begin_0, end = var_2699_end_0, end_mask = var_2699_end_mask_0, x = var_2665_cast)[name = tensor<string, []>("op_2699_cast")];
            tensor<int32, [4]> var_2700_begin_0 = const()[name = tensor<string, []>("op_2700_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2700_end_0 = const()[name = tensor<string, []>("op_2700_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2700_end_mask_0 = const()[name = tensor<string, []>("op_2700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2700_cast = slice_by_index(begin = var_2700_begin_0, end = var_2700_end_0, end_mask = var_2700_end_mask_0, x = var_2669_cast)[name = tensor<string, []>("op_2700_cast")];
            tensor<int32, [4]> var_2701_begin_0 = const()[name = tensor<string, []>("op_2701_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2701_end_0 = const()[name = tensor<string, []>("op_2701_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2701_end_mask_0 = const()[name = tensor<string, []>("op_2701_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2701_cast = slice_by_index(begin = var_2701_begin_0, end = var_2701_end_0, end_mask = var_2701_end_mask_0, x = var_2669_cast)[name = tensor<string, []>("op_2701_cast")];
            tensor<int32, [4]> var_2702_begin_0 = const()[name = tensor<string, []>("op_2702_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2702_end_0 = const()[name = tensor<string, []>("op_2702_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2702_end_mask_0 = const()[name = tensor<string, []>("op_2702_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2702_cast = slice_by_index(begin = var_2702_begin_0, end = var_2702_end_0, end_mask = var_2702_end_mask_0, x = var_2673_cast)[name = tensor<string, []>("op_2702_cast")];
            tensor<int32, [4]> var_2703_begin_0 = const()[name = tensor<string, []>("op_2703_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2703_end_0 = const()[name = tensor<string, []>("op_2703_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2703_end_mask_0 = const()[name = tensor<string, []>("op_2703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2703_cast = slice_by_index(begin = var_2703_begin_0, end = var_2703_end_0, end_mask = var_2703_end_mask_0, x = var_2673_cast)[name = tensor<string, []>("op_2703_cast")];
            tensor<int32, [4]> var_2704_begin_0 = const()[name = tensor<string, []>("op_2704_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2704_end_0 = const()[name = tensor<string, []>("op_2704_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2704_end_mask_0 = const()[name = tensor<string, []>("op_2704_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2704_cast = slice_by_index(begin = var_2704_begin_0, end = var_2704_end_0, end_mask = var_2704_end_mask_0, x = var_2677_cast)[name = tensor<string, []>("op_2704_cast")];
            tensor<int32, [4]> var_2705_begin_0 = const()[name = tensor<string, []>("op_2705_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2705_end_0 = const()[name = tensor<string, []>("op_2705_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2705_end_mask_0 = const()[name = tensor<string, []>("op_2705_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2705_cast = slice_by_index(begin = var_2705_begin_0, end = var_2705_end_0, end_mask = var_2705_end_mask_0, x = var_2677_cast)[name = tensor<string, []>("op_2705_cast")];
            tensor<int32, [4]> var_2706_begin_0 = const()[name = tensor<string, []>("op_2706_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2706_end_0 = const()[name = tensor<string, []>("op_2706_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2706_end_mask_0 = const()[name = tensor<string, []>("op_2706_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2706_cast = slice_by_index(begin = var_2706_begin_0, end = var_2706_end_0, end_mask = var_2706_end_mask_0, x = var_2681_cast)[name = tensor<string, []>("op_2706_cast")];
            tensor<int32, [4]> var_2707_begin_0 = const()[name = tensor<string, []>("op_2707_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2707_end_0 = const()[name = tensor<string, []>("op_2707_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2707_end_mask_0 = const()[name = tensor<string, []>("op_2707_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2707_cast = slice_by_index(begin = var_2707_begin_0, end = var_2707_end_0, end_mask = var_2707_end_mask_0, x = var_2681_cast)[name = tensor<string, []>("op_2707_cast")];
            tensor<int32, [4]> var_2708_begin_0 = const()[name = tensor<string, []>("op_2708_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2708_end_0 = const()[name = tensor<string, []>("op_2708_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2708_end_mask_0 = const()[name = tensor<string, []>("op_2708_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2708_cast = slice_by_index(begin = var_2708_begin_0, end = var_2708_end_0, end_mask = var_2708_end_mask_0, x = var_2685_cast)[name = tensor<string, []>("op_2708_cast")];
            tensor<int32, [4]> var_2709_begin_0 = const()[name = tensor<string, []>("op_2709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2709_end_0 = const()[name = tensor<string, []>("op_2709_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2709_end_mask_0 = const()[name = tensor<string, []>("op_2709_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2709_cast = slice_by_index(begin = var_2709_begin_0, end = var_2709_end_0, end_mask = var_2709_end_mask_0, x = var_2685_cast)[name = tensor<string, []>("op_2709_cast")];
            tensor<int32, [4]> var_2710_begin_0 = const()[name = tensor<string, []>("op_2710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2710_end_0 = const()[name = tensor<string, []>("op_2710_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_2710_end_mask_0 = const()[name = tensor<string, []>("op_2710_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2710_cast = slice_by_index(begin = var_2710_begin_0, end = var_2710_end_0, end_mask = var_2710_end_mask_0, x = var_2689_cast)[name = tensor<string, []>("op_2710_cast")];
            tensor<int32, [4]> var_2711_begin_0 = const()[name = tensor<string, []>("op_2711_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2711_end_0 = const()[name = tensor<string, []>("op_2711_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_2711_end_mask_0 = const()[name = tensor<string, []>("op_2711_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_2711_cast = slice_by_index(begin = var_2711_begin_0, end = var_2711_end_0, end_mask = var_2711_end_mask_0, x = var_2689_cast)[name = tensor<string, []>("op_2711_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_2716_begin_0 = const()[name = tensor<string, []>("op_2716_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2716_end_0 = const()[name = tensor<string, []>("op_2716_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_2716_end_mask_0 = const()[name = tensor<string, []>("op_2716_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, 64]> var_2716_cast = slice_by_index(begin = var_2716_begin_0, end = var_2716_end_0, end_mask = var_2716_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2716_cast")];
            tensor<int32, [4]> var_2720_begin_0 = const()[name = tensor<string, []>("op_2720_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_2720_end_0 = const()[name = tensor<string, []>("op_2720_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_2720_end_mask_0 = const()[name = tensor<string, []>("op_2720_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2720_cast = slice_by_index(begin = var_2720_begin_0, end = var_2720_end_0, end_mask = var_2720_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2720_cast")];
            tensor<int32, [4]> var_2724_begin_0 = const()[name = tensor<string, []>("op_2724_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_2724_end_0 = const()[name = tensor<string, []>("op_2724_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_2724_end_mask_0 = const()[name = tensor<string, []>("op_2724_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2724_cast = slice_by_index(begin = var_2724_begin_0, end = var_2724_end_0, end_mask = var_2724_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2724_cast")];
            tensor<int32, [4]> var_2728_begin_0 = const()[name = tensor<string, []>("op_2728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_2728_end_0 = const()[name = tensor<string, []>("op_2728_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_2728_end_mask_0 = const()[name = tensor<string, []>("op_2728_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2728_cast = slice_by_index(begin = var_2728_begin_0, end = var_2728_end_0, end_mask = var_2728_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2728_cast")];
            tensor<int32, [4]> var_2732_begin_0 = const()[name = tensor<string, []>("op_2732_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2732_end_0 = const()[name = tensor<string, []>("op_2732_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_2732_end_mask_0 = const()[name = tensor<string, []>("op_2732_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2732_cast = slice_by_index(begin = var_2732_begin_0, end = var_2732_end_0, end_mask = var_2732_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2732_cast")];
            tensor<int32, [4]> var_2736_begin_0 = const()[name = tensor<string, []>("op_2736_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2736_end_0 = const()[name = tensor<string, []>("op_2736_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_2736_end_mask_0 = const()[name = tensor<string, []>("op_2736_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2736_cast = slice_by_index(begin = var_2736_begin_0, end = var_2736_end_0, end_mask = var_2736_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2736_cast")];
            tensor<int32, [4]> var_2740_begin_0 = const()[name = tensor<string, []>("op_2740_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_2740_end_0 = const()[name = tensor<string, []>("op_2740_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_2740_end_mask_0 = const()[name = tensor<string, []>("op_2740_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2740_cast = slice_by_index(begin = var_2740_begin_0, end = var_2740_end_0, end_mask = var_2740_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2740_cast")];
            tensor<int32, [4]> var_2744_begin_0 = const()[name = tensor<string, []>("op_2744_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_2744_end_0 = const()[name = tensor<string, []>("op_2744_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_2744_end_mask_0 = const()[name = tensor<string, []>("op_2744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2744_cast = slice_by_index(begin = var_2744_begin_0, end = var_2744_end_0, end_mask = var_2744_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2744_cast")];
            tensor<int32, [4]> var_2748_begin_0 = const()[name = tensor<string, []>("op_2748_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_2748_end_0 = const()[name = tensor<string, []>("op_2748_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_2748_end_mask_0 = const()[name = tensor<string, []>("op_2748_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2748_cast = slice_by_index(begin = var_2748_begin_0, end = var_2748_end_0, end_mask = var_2748_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2748_cast")];
            tensor<int32, [4]> var_2752_begin_0 = const()[name = tensor<string, []>("op_2752_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_2752_end_0 = const()[name = tensor<string, []>("op_2752_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_2752_end_mask_0 = const()[name = tensor<string, []>("op_2752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_2752_cast = slice_by_index(begin = var_2752_begin_0, end = var_2752_end_0, end_mask = var_2752_end_mask_0, x = transpose_26)[name = tensor<string, []>("op_2752_cast")];
            tensor<int32, [4]> var_2754_begin_0 = const()[name = tensor<string, []>("op_2754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2754_end_0 = const()[name = tensor<string, []>("op_2754_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_2754_end_mask_0 = const()[name = tensor<string, []>("op_2754_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2754_cast = slice_by_index(begin = var_2754_begin_0, end = var_2754_end_0, end_mask = var_2754_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2754_cast")];
            tensor<int32, [4]> var_2758_begin_0 = const()[name = tensor<string, []>("op_2758_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2758_end_0 = const()[name = tensor<string, []>("op_2758_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_2758_end_mask_0 = const()[name = tensor<string, []>("op_2758_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2758_cast = slice_by_index(begin = var_2758_begin_0, end = var_2758_end_0, end_mask = var_2758_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2758_cast")];
            tensor<int32, [4]> var_2762_begin_0 = const()[name = tensor<string, []>("op_2762_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2762_end_0 = const()[name = tensor<string, []>("op_2762_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_2762_end_mask_0 = const()[name = tensor<string, []>("op_2762_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2762_cast = slice_by_index(begin = var_2762_begin_0, end = var_2762_end_0, end_mask = var_2762_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2762_cast")];
            tensor<int32, [4]> var_2766_begin_0 = const()[name = tensor<string, []>("op_2766_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2766_end_0 = const()[name = tensor<string, []>("op_2766_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_2766_end_mask_0 = const()[name = tensor<string, []>("op_2766_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2766_cast = slice_by_index(begin = var_2766_begin_0, end = var_2766_end_0, end_mask = var_2766_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2766_cast")];
            tensor<int32, [4]> var_2770_begin_0 = const()[name = tensor<string, []>("op_2770_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2770_end_0 = const()[name = tensor<string, []>("op_2770_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_2770_end_mask_0 = const()[name = tensor<string, []>("op_2770_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2770_cast = slice_by_index(begin = var_2770_begin_0, end = var_2770_end_0, end_mask = var_2770_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2770_cast")];
            tensor<int32, [4]> var_2774_begin_0 = const()[name = tensor<string, []>("op_2774_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2774_end_0 = const()[name = tensor<string, []>("op_2774_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_2774_end_mask_0 = const()[name = tensor<string, []>("op_2774_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2774_cast = slice_by_index(begin = var_2774_begin_0, end = var_2774_end_0, end_mask = var_2774_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2774_cast")];
            tensor<int32, [4]> var_2778_begin_0 = const()[name = tensor<string, []>("op_2778_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_2778_end_0 = const()[name = tensor<string, []>("op_2778_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_2778_end_mask_0 = const()[name = tensor<string, []>("op_2778_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2778_cast = slice_by_index(begin = var_2778_begin_0, end = var_2778_end_0, end_mask = var_2778_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2778_cast")];
            tensor<int32, [4]> var_2782_begin_0 = const()[name = tensor<string, []>("op_2782_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_2782_end_0 = const()[name = tensor<string, []>("op_2782_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_2782_end_mask_0 = const()[name = tensor<string, []>("op_2782_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2782_cast = slice_by_index(begin = var_2782_begin_0, end = var_2782_end_0, end_mask = var_2782_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2782_cast")];
            tensor<int32, [4]> var_2786_begin_0 = const()[name = tensor<string, []>("op_2786_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_2786_end_0 = const()[name = tensor<string, []>("op_2786_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_2786_end_mask_0 = const()[name = tensor<string, []>("op_2786_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2786_cast = slice_by_index(begin = var_2786_begin_0, end = var_2786_end_0, end_mask = var_2786_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2786_cast")];
            tensor<int32, [4]> var_2790_begin_0 = const()[name = tensor<string, []>("op_2790_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_2790_end_0 = const()[name = tensor<string, []>("op_2790_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_2790_end_mask_0 = const()[name = tensor<string, []>("op_2790_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_2790_cast = slice_by_index(begin = var_2790_begin_0, end = var_2790_end_0, end_mask = var_2790_end_mask_0, x = v_11_cast)[name = tensor<string, []>("op_2790_cast")];
            tensor<string, []> var_2794_equation_0 = const()[name = tensor<string, []>("op_2794_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2794_cast = einsum(equation = var_2794_equation_0, values = (var_2716_cast, var_2692_cast))[name = tensor<string, []>("op_2794_cast")];
            tensor<fp16, []> var_2795_to_fp16 = const()[name = tensor<string, []>("op_2795_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_361_cast = mul(x = var_2794_cast, y = var_2795_to_fp16)[name = tensor<string, []>("aw_chunk_361_cast")];
            tensor<string, []> var_2798_equation_0 = const()[name = tensor<string, []>("op_2798_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2798_cast = einsum(equation = var_2798_equation_0, values = (var_2716_cast, var_2693_cast))[name = tensor<string, []>("op_2798_cast")];
            tensor<fp16, []> var_2799_to_fp16 = const()[name = tensor<string, []>("op_2799_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_363_cast = mul(x = var_2798_cast, y = var_2799_to_fp16)[name = tensor<string, []>("aw_chunk_363_cast")];
            tensor<string, []> var_2802_equation_0 = const()[name = tensor<string, []>("op_2802_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2802_cast = einsum(equation = var_2802_equation_0, values = (var_2720_cast, var_2694_cast))[name = tensor<string, []>("op_2802_cast")];
            tensor<fp16, []> var_2803_to_fp16 = const()[name = tensor<string, []>("op_2803_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_365_cast = mul(x = var_2802_cast, y = var_2803_to_fp16)[name = tensor<string, []>("aw_chunk_365_cast")];
            tensor<string, []> var_2806_equation_0 = const()[name = tensor<string, []>("op_2806_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2806_cast = einsum(equation = var_2806_equation_0, values = (var_2720_cast, var_2695_cast))[name = tensor<string, []>("op_2806_cast")];
            tensor<fp16, []> var_2807_to_fp16 = const()[name = tensor<string, []>("op_2807_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_367_cast = mul(x = var_2806_cast, y = var_2807_to_fp16)[name = tensor<string, []>("aw_chunk_367_cast")];
            tensor<string, []> var_2810_equation_0 = const()[name = tensor<string, []>("op_2810_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2810_cast = einsum(equation = var_2810_equation_0, values = (var_2724_cast, var_2696_cast))[name = tensor<string, []>("op_2810_cast")];
            tensor<fp16, []> var_2811_to_fp16 = const()[name = tensor<string, []>("op_2811_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_369_cast = mul(x = var_2810_cast, y = var_2811_to_fp16)[name = tensor<string, []>("aw_chunk_369_cast")];
            tensor<string, []> var_2814_equation_0 = const()[name = tensor<string, []>("op_2814_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2814_cast = einsum(equation = var_2814_equation_0, values = (var_2724_cast, var_2697_cast))[name = tensor<string, []>("op_2814_cast")];
            tensor<fp16, []> var_2815_to_fp16 = const()[name = tensor<string, []>("op_2815_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_371_cast = mul(x = var_2814_cast, y = var_2815_to_fp16)[name = tensor<string, []>("aw_chunk_371_cast")];
            tensor<string, []> var_2818_equation_0 = const()[name = tensor<string, []>("op_2818_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2818_cast = einsum(equation = var_2818_equation_0, values = (var_2728_cast, var_2698_cast))[name = tensor<string, []>("op_2818_cast")];
            tensor<fp16, []> var_2819_to_fp16 = const()[name = tensor<string, []>("op_2819_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_373_cast = mul(x = var_2818_cast, y = var_2819_to_fp16)[name = tensor<string, []>("aw_chunk_373_cast")];
            tensor<string, []> var_2822_equation_0 = const()[name = tensor<string, []>("op_2822_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2822_cast = einsum(equation = var_2822_equation_0, values = (var_2728_cast, var_2699_cast))[name = tensor<string, []>("op_2822_cast")];
            tensor<fp16, []> var_2823_to_fp16 = const()[name = tensor<string, []>("op_2823_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_375_cast = mul(x = var_2822_cast, y = var_2823_to_fp16)[name = tensor<string, []>("aw_chunk_375_cast")];
            tensor<string, []> var_2826_equation_0 = const()[name = tensor<string, []>("op_2826_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2826_cast = einsum(equation = var_2826_equation_0, values = (var_2732_cast, var_2700_cast))[name = tensor<string, []>("op_2826_cast")];
            tensor<fp16, []> var_2827_to_fp16 = const()[name = tensor<string, []>("op_2827_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_377_cast = mul(x = var_2826_cast, y = var_2827_to_fp16)[name = tensor<string, []>("aw_chunk_377_cast")];
            tensor<string, []> var_2830_equation_0 = const()[name = tensor<string, []>("op_2830_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2830_cast = einsum(equation = var_2830_equation_0, values = (var_2732_cast, var_2701_cast))[name = tensor<string, []>("op_2830_cast")];
            tensor<fp16, []> var_2831_to_fp16 = const()[name = tensor<string, []>("op_2831_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_379_cast = mul(x = var_2830_cast, y = var_2831_to_fp16)[name = tensor<string, []>("aw_chunk_379_cast")];
            tensor<string, []> var_2834_equation_0 = const()[name = tensor<string, []>("op_2834_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2834_cast = einsum(equation = var_2834_equation_0, values = (var_2736_cast, var_2702_cast))[name = tensor<string, []>("op_2834_cast")];
            tensor<fp16, []> var_2835_to_fp16 = const()[name = tensor<string, []>("op_2835_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_381_cast = mul(x = var_2834_cast, y = var_2835_to_fp16)[name = tensor<string, []>("aw_chunk_381_cast")];
            tensor<string, []> var_2838_equation_0 = const()[name = tensor<string, []>("op_2838_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2838_cast = einsum(equation = var_2838_equation_0, values = (var_2736_cast, var_2703_cast))[name = tensor<string, []>("op_2838_cast")];
            tensor<fp16, []> var_2839_to_fp16 = const()[name = tensor<string, []>("op_2839_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_383_cast = mul(x = var_2838_cast, y = var_2839_to_fp16)[name = tensor<string, []>("aw_chunk_383_cast")];
            tensor<string, []> var_2842_equation_0 = const()[name = tensor<string, []>("op_2842_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2842_cast = einsum(equation = var_2842_equation_0, values = (var_2740_cast, var_2704_cast))[name = tensor<string, []>("op_2842_cast")];
            tensor<fp16, []> var_2843_to_fp16 = const()[name = tensor<string, []>("op_2843_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_385_cast = mul(x = var_2842_cast, y = var_2843_to_fp16)[name = tensor<string, []>("aw_chunk_385_cast")];
            tensor<string, []> var_2846_equation_0 = const()[name = tensor<string, []>("op_2846_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2846_cast = einsum(equation = var_2846_equation_0, values = (var_2740_cast, var_2705_cast))[name = tensor<string, []>("op_2846_cast")];
            tensor<fp16, []> var_2847_to_fp16 = const()[name = tensor<string, []>("op_2847_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_387_cast = mul(x = var_2846_cast, y = var_2847_to_fp16)[name = tensor<string, []>("aw_chunk_387_cast")];
            tensor<string, []> var_2850_equation_0 = const()[name = tensor<string, []>("op_2850_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2850_cast = einsum(equation = var_2850_equation_0, values = (var_2744_cast, var_2706_cast))[name = tensor<string, []>("op_2850_cast")];
            tensor<fp16, []> var_2851_to_fp16 = const()[name = tensor<string, []>("op_2851_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_389_cast = mul(x = var_2850_cast, y = var_2851_to_fp16)[name = tensor<string, []>("aw_chunk_389_cast")];
            tensor<string, []> var_2854_equation_0 = const()[name = tensor<string, []>("op_2854_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2854_cast = einsum(equation = var_2854_equation_0, values = (var_2744_cast, var_2707_cast))[name = tensor<string, []>("op_2854_cast")];
            tensor<fp16, []> var_2855_to_fp16 = const()[name = tensor<string, []>("op_2855_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_391_cast = mul(x = var_2854_cast, y = var_2855_to_fp16)[name = tensor<string, []>("aw_chunk_391_cast")];
            tensor<string, []> var_2858_equation_0 = const()[name = tensor<string, []>("op_2858_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2858_cast = einsum(equation = var_2858_equation_0, values = (var_2748_cast, var_2708_cast))[name = tensor<string, []>("op_2858_cast")];
            tensor<fp16, []> var_2859_to_fp16 = const()[name = tensor<string, []>("op_2859_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_393_cast = mul(x = var_2858_cast, y = var_2859_to_fp16)[name = tensor<string, []>("aw_chunk_393_cast")];
            tensor<string, []> var_2862_equation_0 = const()[name = tensor<string, []>("op_2862_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2862_cast = einsum(equation = var_2862_equation_0, values = (var_2748_cast, var_2709_cast))[name = tensor<string, []>("op_2862_cast")];
            tensor<fp16, []> var_2863_to_fp16 = const()[name = tensor<string, []>("op_2863_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_395_cast = mul(x = var_2862_cast, y = var_2863_to_fp16)[name = tensor<string, []>("aw_chunk_395_cast")];
            tensor<string, []> var_2866_equation_0 = const()[name = tensor<string, []>("op_2866_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2866_cast = einsum(equation = var_2866_equation_0, values = (var_2752_cast, var_2710_cast))[name = tensor<string, []>("op_2866_cast")];
            tensor<fp16, []> var_2867_to_fp16 = const()[name = tensor<string, []>("op_2867_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_397_cast = mul(x = var_2866_cast, y = var_2867_to_fp16)[name = tensor<string, []>("aw_chunk_397_cast")];
            tensor<string, []> var_2870_equation_0 = const()[name = tensor<string, []>("op_2870_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_2870_cast = einsum(equation = var_2870_equation_0, values = (var_2752_cast, var_2711_cast))[name = tensor<string, []>("op_2870_cast")];
            tensor<fp16, []> var_2871_to_fp16 = const()[name = tensor<string, []>("op_2871_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_399_cast = mul(x = var_2870_cast, y = var_2871_to_fp16)[name = tensor<string, []>("aw_chunk_399_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2873_cast = softmax(axis = var_2173, x = aw_chunk_361_cast)[name = tensor<string, []>("op_2873_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2874_cast = softmax(axis = var_2173, x = aw_chunk_363_cast)[name = tensor<string, []>("op_2874_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2875_cast = softmax(axis = var_2173, x = aw_chunk_365_cast)[name = tensor<string, []>("op_2875_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2876_cast = softmax(axis = var_2173, x = aw_chunk_367_cast)[name = tensor<string, []>("op_2876_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2877_cast = softmax(axis = var_2173, x = aw_chunk_369_cast)[name = tensor<string, []>("op_2877_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2878_cast = softmax(axis = var_2173, x = aw_chunk_371_cast)[name = tensor<string, []>("op_2878_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2879_cast = softmax(axis = var_2173, x = aw_chunk_373_cast)[name = tensor<string, []>("op_2879_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2880_cast = softmax(axis = var_2173, x = aw_chunk_375_cast)[name = tensor<string, []>("op_2880_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2881_cast = softmax(axis = var_2173, x = aw_chunk_377_cast)[name = tensor<string, []>("op_2881_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2882_cast = softmax(axis = var_2173, x = aw_chunk_379_cast)[name = tensor<string, []>("op_2882_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2883_cast = softmax(axis = var_2173, x = aw_chunk_381_cast)[name = tensor<string, []>("op_2883_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2884_cast = softmax(axis = var_2173, x = aw_chunk_383_cast)[name = tensor<string, []>("op_2884_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2885_cast = softmax(axis = var_2173, x = aw_chunk_385_cast)[name = tensor<string, []>("op_2885_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2886_cast = softmax(axis = var_2173, x = aw_chunk_387_cast)[name = tensor<string, []>("op_2886_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2887_cast = softmax(axis = var_2173, x = aw_chunk_389_cast)[name = tensor<string, []>("op_2887_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2888_cast = softmax(axis = var_2173, x = aw_chunk_391_cast)[name = tensor<string, []>("op_2888_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2889_cast = softmax(axis = var_2173, x = aw_chunk_393_cast)[name = tensor<string, []>("op_2889_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2890_cast = softmax(axis = var_2173, x = aw_chunk_395_cast)[name = tensor<string, []>("op_2890_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2891_cast = softmax(axis = var_2173, x = aw_chunk_397_cast)[name = tensor<string, []>("op_2891_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_2892_cast = softmax(axis = var_2173, x = aw_chunk_399_cast)[name = tensor<string, []>("op_2892_cast")];
            tensor<string, []> var_2894_equation_0 = const()[name = tensor<string, []>("op_2894_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2894_cast = einsum(equation = var_2894_equation_0, values = (var_2754_cast, var_2873_cast))[name = tensor<string, []>("op_2894_cast")];
            tensor<string, []> var_2896_equation_0 = const()[name = tensor<string, []>("op_2896_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2896_cast = einsum(equation = var_2896_equation_0, values = (var_2754_cast, var_2874_cast))[name = tensor<string, []>("op_2896_cast")];
            tensor<string, []> var_2898_equation_0 = const()[name = tensor<string, []>("op_2898_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2898_cast = einsum(equation = var_2898_equation_0, values = (var_2758_cast, var_2875_cast))[name = tensor<string, []>("op_2898_cast")];
            tensor<string, []> var_2900_equation_0 = const()[name = tensor<string, []>("op_2900_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2900_cast = einsum(equation = var_2900_equation_0, values = (var_2758_cast, var_2876_cast))[name = tensor<string, []>("op_2900_cast")];
            tensor<string, []> var_2902_equation_0 = const()[name = tensor<string, []>("op_2902_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2902_cast = einsum(equation = var_2902_equation_0, values = (var_2762_cast, var_2877_cast))[name = tensor<string, []>("op_2902_cast")];
            tensor<string, []> var_2904_equation_0 = const()[name = tensor<string, []>("op_2904_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2904_cast = einsum(equation = var_2904_equation_0, values = (var_2762_cast, var_2878_cast))[name = tensor<string, []>("op_2904_cast")];
            tensor<string, []> var_2906_equation_0 = const()[name = tensor<string, []>("op_2906_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2906_cast = einsum(equation = var_2906_equation_0, values = (var_2766_cast, var_2879_cast))[name = tensor<string, []>("op_2906_cast")];
            tensor<string, []> var_2908_equation_0 = const()[name = tensor<string, []>("op_2908_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2908_cast = einsum(equation = var_2908_equation_0, values = (var_2766_cast, var_2880_cast))[name = tensor<string, []>("op_2908_cast")];
            tensor<string, []> var_2910_equation_0 = const()[name = tensor<string, []>("op_2910_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2910_cast = einsum(equation = var_2910_equation_0, values = (var_2770_cast, var_2881_cast))[name = tensor<string, []>("op_2910_cast")];
            tensor<string, []> var_2912_equation_0 = const()[name = tensor<string, []>("op_2912_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2912_cast = einsum(equation = var_2912_equation_0, values = (var_2770_cast, var_2882_cast))[name = tensor<string, []>("op_2912_cast")];
            tensor<string, []> var_2914_equation_0 = const()[name = tensor<string, []>("op_2914_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2914_cast = einsum(equation = var_2914_equation_0, values = (var_2774_cast, var_2883_cast))[name = tensor<string, []>("op_2914_cast")];
            tensor<string, []> var_2916_equation_0 = const()[name = tensor<string, []>("op_2916_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2916_cast = einsum(equation = var_2916_equation_0, values = (var_2774_cast, var_2884_cast))[name = tensor<string, []>("op_2916_cast")];
            tensor<string, []> var_2918_equation_0 = const()[name = tensor<string, []>("op_2918_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2918_cast = einsum(equation = var_2918_equation_0, values = (var_2778_cast, var_2885_cast))[name = tensor<string, []>("op_2918_cast")];
            tensor<string, []> var_2920_equation_0 = const()[name = tensor<string, []>("op_2920_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2920_cast = einsum(equation = var_2920_equation_0, values = (var_2778_cast, var_2886_cast))[name = tensor<string, []>("op_2920_cast")];
            tensor<string, []> var_2922_equation_0 = const()[name = tensor<string, []>("op_2922_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2922_cast = einsum(equation = var_2922_equation_0, values = (var_2782_cast, var_2887_cast))[name = tensor<string, []>("op_2922_cast")];
            tensor<string, []> var_2924_equation_0 = const()[name = tensor<string, []>("op_2924_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2924_cast = einsum(equation = var_2924_equation_0, values = (var_2782_cast, var_2888_cast))[name = tensor<string, []>("op_2924_cast")];
            tensor<string, []> var_2926_equation_0 = const()[name = tensor<string, []>("op_2926_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2926_cast = einsum(equation = var_2926_equation_0, values = (var_2786_cast, var_2889_cast))[name = tensor<string, []>("op_2926_cast")];
            tensor<string, []> var_2928_equation_0 = const()[name = tensor<string, []>("op_2928_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2928_cast = einsum(equation = var_2928_equation_0, values = (var_2786_cast, var_2890_cast))[name = tensor<string, []>("op_2928_cast")];
            tensor<string, []> var_2930_equation_0 = const()[name = tensor<string, []>("op_2930_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2930_cast = einsum(equation = var_2930_equation_0, values = (var_2790_cast, var_2891_cast))[name = tensor<string, []>("op_2930_cast")];
            tensor<string, []> var_2932_equation_0 = const()[name = tensor<string, []>("op_2932_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_2932_cast = einsum(equation = var_2932_equation_0, values = (var_2790_cast, var_2892_cast))[name = tensor<string, []>("op_2932_cast")];
            tensor<bool, []> var_2934_interleave_0 = const()[name = tensor<string, []>("op_2934_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2934_cast = concat(axis = var_2150, interleave = var_2934_interleave_0, values = (var_2894_cast, var_2896_cast))[name = tensor<string, []>("op_2934_cast")];
            tensor<bool, []> var_2936_interleave_0 = const()[name = tensor<string, []>("op_2936_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2936_cast = concat(axis = var_2150, interleave = var_2936_interleave_0, values = (var_2898_cast, var_2900_cast))[name = tensor<string, []>("op_2936_cast")];
            tensor<bool, []> var_2938_interleave_0 = const()[name = tensor<string, []>("op_2938_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2938_cast = concat(axis = var_2150, interleave = var_2938_interleave_0, values = (var_2902_cast, var_2904_cast))[name = tensor<string, []>("op_2938_cast")];
            tensor<bool, []> var_2940_interleave_0 = const()[name = tensor<string, []>("op_2940_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2940_cast = concat(axis = var_2150, interleave = var_2940_interleave_0, values = (var_2906_cast, var_2908_cast))[name = tensor<string, []>("op_2940_cast")];
            tensor<bool, []> var_2942_interleave_0 = const()[name = tensor<string, []>("op_2942_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2942_cast = concat(axis = var_2150, interleave = var_2942_interleave_0, values = (var_2910_cast, var_2912_cast))[name = tensor<string, []>("op_2942_cast")];
            tensor<bool, []> var_2944_interleave_0 = const()[name = tensor<string, []>("op_2944_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2944_cast = concat(axis = var_2150, interleave = var_2944_interleave_0, values = (var_2914_cast, var_2916_cast))[name = tensor<string, []>("op_2944_cast")];
            tensor<bool, []> var_2946_interleave_0 = const()[name = tensor<string, []>("op_2946_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2946_cast = concat(axis = var_2150, interleave = var_2946_interleave_0, values = (var_2918_cast, var_2920_cast))[name = tensor<string, []>("op_2946_cast")];
            tensor<bool, []> var_2948_interleave_0 = const()[name = tensor<string, []>("op_2948_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2948_cast = concat(axis = var_2150, interleave = var_2948_interleave_0, values = (var_2922_cast, var_2924_cast))[name = tensor<string, []>("op_2948_cast")];
            tensor<bool, []> var_2950_interleave_0 = const()[name = tensor<string, []>("op_2950_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2950_cast = concat(axis = var_2150, interleave = var_2950_interleave_0, values = (var_2926_cast, var_2928_cast))[name = tensor<string, []>("op_2950_cast")];
            tensor<bool, []> var_2952_interleave_0 = const()[name = tensor<string, []>("op_2952_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_2952_cast = concat(axis = var_2150, interleave = var_2952_interleave_0, values = (var_2930_cast, var_2932_cast))[name = tensor<string, []>("op_2952_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_2173, interleave = input_81_interleave_0, values = (var_2934_cast, var_2936_cast, var_2938_cast, var_2940_cast, var_2942_cast, var_2944_cast, var_2946_cast, var_2948_cast, var_2950_cast, var_2952_cast))[name = tensor<string, []>("input_81_cast")];
            tensor<int32, [2]> var_2958 = const()[name = tensor<string, []>("op_2958"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2960 = const()[name = tensor<string, []>("op_2960"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2962_pad_type_0 = const()[name = tensor<string, []>("op_2962_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2962_pad_0 = const()[name = tensor<string, []>("op_2962_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5867776))), lut = tensor<fp16, [4]>([-0x1.894p-6, -0x1.b64p-8, 0x1.bbp-8, 0x1.8ap-6]), 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, []>(5970240)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2962_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_2960, groups = var_2173, pad = var_2962_pad_0, pad_type = var_2962_pad_type_0, strides = var_2958, 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_2962_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_17_cast = add(x = var_2962_cast, y = inputs_15_cast)[name = tensor<string, []>("inputs_17_cast")];
            tensor<int32, [1]> var_2966 = const()[name = tensor<string, []>("op_2966"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_17_cast = reduce_mean(axes = var_2966, keep_dims = var_2168, 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_2970 = const()[name = tensor<string, []>("op_2970"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_2971_cast = reduce_mean(axes = var_2970, keep_dims = var_2168, x = zero_mean_sq_17_cast)[name = tensor<string, []>("op_2971_cast")];
            tensor<fp16, []> var_2972_to_fp16 = const()[name = tensor<string, []>("op_2972_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_2973_cast = add(x = var_2971_cast, y = var_2972_to_fp16)[name = tensor<string, []>("op_2973_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_2973_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_2977_to_fp16 = const()[name = tensor<string, []>("op_2977_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5971584)))];
            tensor<fp16, [2, 640, 1, 1024]> var_2978_cast = add(x = out_17_cast, y = var_2977_to_fp16)[name = tensor<string, []>("op_2978_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_2980_to_fp16 = const()[name = tensor<string, []>("op_2980_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5972928)))];
            tensor<fp16, [2, 640, 1, 1024]> input_83_cast = mul(x = var_2978_cast, y = var_2980_to_fp16)[name = tensor<string, []>("input_83_cast")];
            tensor<int32, [2]> var_2988 = const()[name = tensor<string, []>("op_2988"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_2990 = const()[name = tensor<string, []>("op_2990"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_2992_pad_type_0 = const()[name = tensor<string, []>("op_2992_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_2992_pad_0 = const()[name = tensor<string, []>("op_2992_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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5974272))), lut = tensor<fp16, [4]>([-0x1.8acp-4, -0x1.cccp-6, 0x1.d64p-6, 0x1.8ecp-4]), 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6793536))), lut = tensor<fp16, [4]>([0x1.3ap-5, -0x1.79cp-4, -0x1.e2cp-3, -0x1.298p-6]), 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_2992_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_2990, groups = var_2173, pad = var_2992_pad_0, pad_type = var_2992_pad_type_0, strides = var_2988, 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_2992_cast")];
            tensor<int32, [2]> var_2993_split_sizes_0 = const()[name = tensor<string, []>("op_2993_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_2993_axis_0 = const()[name = tensor<string, []>("op_2993_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_2993_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_2993_cast_1 = split(axis = var_2993_axis_0, split_sizes = var_2993_split_sizes_0, x = var_2992_cast)[name = tensor<string, []>("op_2993_cast")];
            tensor<string, []> var_2995_mode_0 = const()[name = tensor<string, []>("op_2995_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_2995_cast = gelu(mode = var_2995_mode_0, x = var_2993_cast_1)[name = tensor<string, []>("op_2995_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_85_cast = mul(x = var_2993_cast_0, y = var_2995_cast)[name = tensor<string, []>("input_85_cast")];
            tensor<int32, [2]> var_2999 = const()[name = tensor<string, []>("op_2999"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3001 = const()[name = tensor<string, []>("op_3001"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3003_pad_type_0 = const()[name = tensor<string, []>("op_3003_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3003_pad_0 = const()[name = tensor<string, []>("op_3003_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6794880))), lut = tensor<fp16, [4]>([-0x1.834p-4, -0x1.cdp-6, 0x1.d18p-6, 0x1.854p-4]), 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, []>(7204544)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3003_cast = conv(bias = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3001, groups = var_2173, pad = var_3003_pad_0, pad_type = var_3003_pad_type_0, strides = var_2999, weight = down_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_85_cast)[name = tensor<string, []>("op_3003_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_51_cast = add(x = var_3003_cast, y = inputs_17_cast)[name = tensor<string, []>("hidden_states_51_cast")];
            tensor<int32, [4]> var_3005 = const()[name = tensor<string, []>("op_3005"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_87_cast = reshape(shape = var_3005, x = hidden_states_51_cast)[name = tensor<string, []>("input_87_cast")];
            tensor<int32, [2]> var_3009 = const()[name = tensor<string, []>("op_3009"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3011 = const()[name = tensor<string, []>("op_3011"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7205888))), lut = tensor<fp16, [4]>([-0x1.32cp-4, -0x1.6dcp-6, 0x1.7p-6, 0x1.338p-4]), 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, []>(7308352)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_53_cast = conv(bias = down_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_3011, groups = var_2173, pad = hidden_states_53_pad_0, pad_type = hidden_states_53_pad_type_0, strides = var_3009, 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, []>(7309696)))];
            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, []>(7311040)))];
            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_3026 = const()[name = tensor<string, []>("op_3026"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3028 = const()[name = tensor<string, []>("op_3028"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7312384))), lut = tensor<fp16, [4]>([-0x1.b18p-4, -0x1.aa8p-6, 0x1.dap-6, 0x1.cbcp-4]), 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, []>(8234048)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_55_cast = conv(bias = down_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_3028, groups = var_2173, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = var_3026, 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_3034 = const()[name = tensor<string, []>("op_3034"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3036 = const()[name = tensor<string, []>("op_3036"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8235392))), lut = tensor<fp16, [4]>([-0x1.948p-5, -0x1.edp-8, 0x1.7fcp-11, 0x1.36cp-7]), 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, []>(8440256)))];
            tensor<fp16, [2, 640, 1, 1]> temb_7_cast = conv(bias = down_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_3036, groups = var_2173, pad = temb_7_pad_0, pad_type = temb_7_pad_type_0, strides = var_3034, 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, []>(8441600)))];
            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, []>(8442944)))];
            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_3046 = const()[name = tensor<string, []>("op_3046"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3048 = const()[name = tensor<string, []>("op_3048"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8444288))), lut = tensor<fp16, [4]>([-0x1.834p-4, -0x1.88p-6, 0x1.6fp-6, 0x1.778p-4]), 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, []>(9365952)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_57_cast = conv(bias = down_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_3048, groups = var_2173, pad = hidden_states_57_pad_0, pad_type = hidden_states_57_pad_type_0, strides = var_3046, 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, []>(9367296)))];
            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, []>(9368640)))];
            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_3068 = const()[name = tensor<string, []>("op_3068"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3070 = const()[name = tensor<string, []>("op_3070"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9369984))), lut = tensor<fp16, [4]>([-0x1.a58p-4, -0x1.f0cp-6, 0x1.06p-5, 0x1.adcp-4]), 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, []>(9472448)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_61_cast = conv(bias = down_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_3070, groups = var_2173, pad = hidden_states_61_pad_0, pad_type = hidden_states_61_pad_type_0, strides = var_3068, 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_3075 = const()[name = tensor<string, []>("op_3075"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_19_cast = reshape(shape = var_3075, x = hidden_states_61_cast)[name = tensor<string, []>("inputs_19_cast")];
            tensor<int32, [1]> var_3085 = const()[name = tensor<string, []>("op_3085"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_19_cast = reduce_mean(axes = var_3085, keep_dims = var_2168, 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_3089 = const()[name = tensor<string, []>("op_3089"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3090_cast = reduce_mean(axes = var_3089, keep_dims = var_2168, x = zero_mean_sq_19_cast)[name = tensor<string, []>("op_3090_cast")];
            tensor<fp16, []> var_3091_to_fp16 = const()[name = tensor<string, []>("op_3091_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3092_cast = add(x = var_3090_cast, y = var_3091_to_fp16)[name = tensor<string, []>("op_3092_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_3092_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_3096_to_fp16 = const()[name = tensor<string, []>("op_3096_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9473792)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3097_cast = add(x = out_19_cast, y = var_3096_to_fp16)[name = tensor<string, []>("op_3097_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3099_to_fp16 = const()[name = tensor<string, []>("op_3099_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9475136)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_63_cast = mul(x = var_3097_cast, y = var_3099_to_fp16)[name = tensor<string, []>("hidden_states_63_cast")];
            tensor<int32, [2]> var_3106 = const()[name = tensor<string, []>("op_3106"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3108 = const()[name = tensor<string, []>("op_3108"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9476480))), lut = tensor<fp16, [4]>([-0x1.fb4p-4, -0x1.3p-5, 0x1.2acp-5, 0x1.f8p-4]), 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_3108, groups = var_2173, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = var_3106, 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_3112 = const()[name = tensor<string, []>("op_3112"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3114 = const()[name = tensor<string, []>("op_3114"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9578944))), lut = tensor<fp16, [4]>([-0x1.f5p-4, -0x1.264p-5, 0x1.32p-5, 0x1.f9cp-4]), 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_3114, groups = var_2173, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = var_3112, 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_3118 = const()[name = tensor<string, []>("op_3118"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3120 = const()[name = tensor<string, []>("op_3120"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9681408))), lut = tensor<fp16, [4]>([-0x1.5p-4, -0x1.8a4p-6, 0x1.8e4p-6, 0x1.514p-4]), 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_3120, groups = var_2173, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = var_3118, 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_3124_begin_0 = const()[name = tensor<string, []>("op_3124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3124_end_0 = const()[name = tensor<string, []>("op_3124_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3124_end_mask_0 = const()[name = tensor<string, []>("op_3124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3124_cast = slice_by_index(begin = var_3124_begin_0, end = var_3124_end_0, end_mask = var_3124_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3124_cast")];
            tensor<int32, [4]> var_3128_begin_0 = const()[name = tensor<string, []>("op_3128_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3128_end_0 = const()[name = tensor<string, []>("op_3128_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_3128_end_mask_0 = const()[name = tensor<string, []>("op_3128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3128_cast = slice_by_index(begin = var_3128_begin_0, end = var_3128_end_0, end_mask = var_3128_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3128_cast")];
            tensor<int32, [4]> var_3132_begin_0 = const()[name = tensor<string, []>("op_3132_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3132_end_0 = const()[name = tensor<string, []>("op_3132_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_3132_end_mask_0 = const()[name = tensor<string, []>("op_3132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3132_cast = slice_by_index(begin = var_3132_begin_0, end = var_3132_end_0, end_mask = var_3132_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3132_cast")];
            tensor<int32, [4]> var_3136_begin_0 = const()[name = tensor<string, []>("op_3136_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3136_end_0 = const()[name = tensor<string, []>("op_3136_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3136_end_mask_0 = const()[name = tensor<string, []>("op_3136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3136_cast = slice_by_index(begin = var_3136_begin_0, end = var_3136_end_0, end_mask = var_3136_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3136_cast")];
            tensor<int32, [4]> var_3140_begin_0 = const()[name = tensor<string, []>("op_3140_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3140_end_0 = const()[name = tensor<string, []>("op_3140_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3140_end_mask_0 = const()[name = tensor<string, []>("op_3140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3140_cast = slice_by_index(begin = var_3140_begin_0, end = var_3140_end_0, end_mask = var_3140_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3140_cast")];
            tensor<int32, [4]> var_3144_begin_0 = const()[name = tensor<string, []>("op_3144_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3144_end_0 = const()[name = tensor<string, []>("op_3144_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_3144_end_mask_0 = const()[name = tensor<string, []>("op_3144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3144_cast = slice_by_index(begin = var_3144_begin_0, end = var_3144_end_0, end_mask = var_3144_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3144_cast")];
            tensor<int32, [4]> var_3148_begin_0 = const()[name = tensor<string, []>("op_3148_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3148_end_0 = const()[name = tensor<string, []>("op_3148_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_3148_end_mask_0 = const()[name = tensor<string, []>("op_3148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3148_cast = slice_by_index(begin = var_3148_begin_0, end = var_3148_end_0, end_mask = var_3148_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3148_cast")];
            tensor<int32, [4]> var_3152_begin_0 = const()[name = tensor<string, []>("op_3152_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3152_end_0 = const()[name = tensor<string, []>("op_3152_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_3152_end_mask_0 = const()[name = tensor<string, []>("op_3152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3152_cast = slice_by_index(begin = var_3152_begin_0, end = var_3152_end_0, end_mask = var_3152_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3152_cast")];
            tensor<int32, [4]> var_3156_begin_0 = const()[name = tensor<string, []>("op_3156_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3156_end_0 = const()[name = tensor<string, []>("op_3156_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_3156_end_mask_0 = const()[name = tensor<string, []>("op_3156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3156_cast = slice_by_index(begin = var_3156_begin_0, end = var_3156_end_0, end_mask = var_3156_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3156_cast")];
            tensor<int32, [4]> var_3160_begin_0 = const()[name = tensor<string, []>("op_3160_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3160_end_0 = const()[name = tensor<string, []>("op_3160_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3160_end_mask_0 = const()[name = tensor<string, []>("op_3160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3160_cast = slice_by_index(begin = var_3160_begin_0, end = var_3160_end_0, end_mask = var_3160_end_mask_0, x = q_13_cast)[name = tensor<string, []>("op_3160_cast")];
            tensor<int32, [4]> var_3163_begin_0 = const()[name = tensor<string, []>("op_3163_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3163_end_0 = const()[name = tensor<string, []>("op_3163_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3163_end_mask_0 = const()[name = tensor<string, []>("op_3163_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3163_cast = slice_by_index(begin = var_3163_begin_0, end = var_3163_end_0, end_mask = var_3163_end_mask_0, x = var_3124_cast)[name = tensor<string, []>("op_3163_cast")];
            tensor<int32, [4]> var_3164_begin_0 = const()[name = tensor<string, []>("op_3164_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3164_end_0 = const()[name = tensor<string, []>("op_3164_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3164_end_mask_0 = const()[name = tensor<string, []>("op_3164_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3164_cast = slice_by_index(begin = var_3164_begin_0, end = var_3164_end_0, end_mask = var_3164_end_mask_0, x = var_3124_cast)[name = tensor<string, []>("op_3164_cast")];
            tensor<int32, [4]> var_3165_begin_0 = const()[name = tensor<string, []>("op_3165_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3165_end_0 = const()[name = tensor<string, []>("op_3165_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3165_end_mask_0 = const()[name = tensor<string, []>("op_3165_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3165_cast = slice_by_index(begin = var_3165_begin_0, end = var_3165_end_0, end_mask = var_3165_end_mask_0, x = var_3128_cast)[name = tensor<string, []>("op_3165_cast")];
            tensor<int32, [4]> var_3166_begin_0 = const()[name = tensor<string, []>("op_3166_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3166_end_0 = const()[name = tensor<string, []>("op_3166_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3166_end_mask_0 = const()[name = tensor<string, []>("op_3166_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3166_cast = slice_by_index(begin = var_3166_begin_0, end = var_3166_end_0, end_mask = var_3166_end_mask_0, x = var_3128_cast)[name = tensor<string, []>("op_3166_cast")];
            tensor<int32, [4]> var_3167_begin_0 = const()[name = tensor<string, []>("op_3167_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3167_end_0 = const()[name = tensor<string, []>("op_3167_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3167_end_mask_0 = const()[name = tensor<string, []>("op_3167_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3167_cast = slice_by_index(begin = var_3167_begin_0, end = var_3167_end_0, end_mask = var_3167_end_mask_0, x = var_3132_cast)[name = tensor<string, []>("op_3167_cast")];
            tensor<int32, [4]> var_3168_begin_0 = const()[name = tensor<string, []>("op_3168_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3168_end_0 = const()[name = tensor<string, []>("op_3168_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3168_end_mask_0 = const()[name = tensor<string, []>("op_3168_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3168_cast = slice_by_index(begin = var_3168_begin_0, end = var_3168_end_0, end_mask = var_3168_end_mask_0, x = var_3132_cast)[name = tensor<string, []>("op_3168_cast")];
            tensor<int32, [4]> var_3169_begin_0 = const()[name = tensor<string, []>("op_3169_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3169_end_0 = const()[name = tensor<string, []>("op_3169_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3169_end_mask_0 = const()[name = tensor<string, []>("op_3169_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3169_cast = slice_by_index(begin = var_3169_begin_0, end = var_3169_end_0, end_mask = var_3169_end_mask_0, x = var_3136_cast)[name = tensor<string, []>("op_3169_cast")];
            tensor<int32, [4]> var_3170_begin_0 = const()[name = tensor<string, []>("op_3170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3170_end_0 = const()[name = tensor<string, []>("op_3170_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3170_end_mask_0 = const()[name = tensor<string, []>("op_3170_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3170_cast = slice_by_index(begin = var_3170_begin_0, end = var_3170_end_0, end_mask = var_3170_end_mask_0, x = var_3136_cast)[name = tensor<string, []>("op_3170_cast")];
            tensor<int32, [4]> var_3171_begin_0 = const()[name = tensor<string, []>("op_3171_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3171_end_0 = const()[name = tensor<string, []>("op_3171_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3171_end_mask_0 = const()[name = tensor<string, []>("op_3171_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3171_cast = slice_by_index(begin = var_3171_begin_0, end = var_3171_end_0, end_mask = var_3171_end_mask_0, x = var_3140_cast)[name = tensor<string, []>("op_3171_cast")];
            tensor<int32, [4]> var_3172_begin_0 = const()[name = tensor<string, []>("op_3172_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3172_end_0 = const()[name = tensor<string, []>("op_3172_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3172_end_mask_0 = const()[name = tensor<string, []>("op_3172_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3172_cast = slice_by_index(begin = var_3172_begin_0, end = var_3172_end_0, end_mask = var_3172_end_mask_0, x = var_3140_cast)[name = tensor<string, []>("op_3172_cast")];
            tensor<int32, [4]> var_3173_begin_0 = const()[name = tensor<string, []>("op_3173_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3173_end_0 = const()[name = tensor<string, []>("op_3173_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3173_end_mask_0 = const()[name = tensor<string, []>("op_3173_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3173_cast = slice_by_index(begin = var_3173_begin_0, end = var_3173_end_0, end_mask = var_3173_end_mask_0, x = var_3144_cast)[name = tensor<string, []>("op_3173_cast")];
            tensor<int32, [4]> var_3174_begin_0 = const()[name = tensor<string, []>("op_3174_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3174_end_0 = const()[name = tensor<string, []>("op_3174_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3174_end_mask_0 = const()[name = tensor<string, []>("op_3174_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3174_cast = slice_by_index(begin = var_3174_begin_0, end = var_3174_end_0, end_mask = var_3174_end_mask_0, x = var_3144_cast)[name = tensor<string, []>("op_3174_cast")];
            tensor<int32, [4]> var_3175_begin_0 = const()[name = tensor<string, []>("op_3175_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3175_end_0 = const()[name = tensor<string, []>("op_3175_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3175_end_mask_0 = const()[name = tensor<string, []>("op_3175_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3175_cast = slice_by_index(begin = var_3175_begin_0, end = var_3175_end_0, end_mask = var_3175_end_mask_0, x = var_3148_cast)[name = tensor<string, []>("op_3175_cast")];
            tensor<int32, [4]> var_3176_begin_0 = const()[name = tensor<string, []>("op_3176_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3176_end_0 = const()[name = tensor<string, []>("op_3176_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3176_end_mask_0 = const()[name = tensor<string, []>("op_3176_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3176_cast = slice_by_index(begin = var_3176_begin_0, end = var_3176_end_0, end_mask = var_3176_end_mask_0, x = var_3148_cast)[name = tensor<string, []>("op_3176_cast")];
            tensor<int32, [4]> var_3177_begin_0 = const()[name = tensor<string, []>("op_3177_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3177_end_0 = const()[name = tensor<string, []>("op_3177_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3177_end_mask_0 = const()[name = tensor<string, []>("op_3177_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3177_cast = slice_by_index(begin = var_3177_begin_0, end = var_3177_end_0, end_mask = var_3177_end_mask_0, x = var_3152_cast)[name = tensor<string, []>("op_3177_cast")];
            tensor<int32, [4]> var_3178_begin_0 = const()[name = tensor<string, []>("op_3178_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3178_end_0 = const()[name = tensor<string, []>("op_3178_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3178_end_mask_0 = const()[name = tensor<string, []>("op_3178_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3178_cast = slice_by_index(begin = var_3178_begin_0, end = var_3178_end_0, end_mask = var_3178_end_mask_0, x = var_3152_cast)[name = tensor<string, []>("op_3178_cast")];
            tensor<int32, [4]> var_3179_begin_0 = const()[name = tensor<string, []>("op_3179_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3179_end_0 = const()[name = tensor<string, []>("op_3179_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3179_end_mask_0 = const()[name = tensor<string, []>("op_3179_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3179_cast = slice_by_index(begin = var_3179_begin_0, end = var_3179_end_0, end_mask = var_3179_end_mask_0, x = var_3156_cast)[name = tensor<string, []>("op_3179_cast")];
            tensor<int32, [4]> var_3180_begin_0 = const()[name = tensor<string, []>("op_3180_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3180_end_0 = const()[name = tensor<string, []>("op_3180_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3180_end_mask_0 = const()[name = tensor<string, []>("op_3180_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3180_cast = slice_by_index(begin = var_3180_begin_0, end = var_3180_end_0, end_mask = var_3180_end_mask_0, x = var_3156_cast)[name = tensor<string, []>("op_3180_cast")];
            tensor<int32, [4]> var_3181_begin_0 = const()[name = tensor<string, []>("op_3181_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3181_end_0 = const()[name = tensor<string, []>("op_3181_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3181_end_mask_0 = const()[name = tensor<string, []>("op_3181_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3181_cast = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = var_3160_cast)[name = tensor<string, []>("op_3181_cast")];
            tensor<int32, [4]> var_3182_begin_0 = const()[name = tensor<string, []>("op_3182_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3182_end_0 = const()[name = tensor<string, []>("op_3182_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3182_end_mask_0 = const()[name = tensor<string, []>("op_3182_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3182_cast = slice_by_index(begin = var_3182_begin_0, end = var_3182_end_0, end_mask = var_3182_end_mask_0, x = var_3160_cast)[name = tensor<string, []>("op_3182_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_3187_begin_0 = const()[name = tensor<string, []>("op_3187_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3187_end_0 = const()[name = tensor<string, []>("op_3187_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_3187_end_mask_0 = const()[name = tensor<string, []>("op_3187_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, 64]> var_3187_cast = slice_by_index(begin = var_3187_begin_0, end = var_3187_end_0, end_mask = var_3187_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3187_cast")];
            tensor<int32, [4]> var_3191_begin_0 = const()[name = tensor<string, []>("op_3191_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_3191_end_0 = const()[name = tensor<string, []>("op_3191_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_3191_end_mask_0 = const()[name = tensor<string, []>("op_3191_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3191_cast = slice_by_index(begin = var_3191_begin_0, end = var_3191_end_0, end_mask = var_3191_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3191_cast")];
            tensor<int32, [4]> var_3195_begin_0 = const()[name = tensor<string, []>("op_3195_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_3195_end_0 = const()[name = tensor<string, []>("op_3195_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_3195_end_mask_0 = const()[name = tensor<string, []>("op_3195_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3195_cast = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3195_cast")];
            tensor<int32, [4]> var_3199_begin_0 = const()[name = tensor<string, []>("op_3199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_3199_end_0 = const()[name = tensor<string, []>("op_3199_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_3199_end_mask_0 = const()[name = tensor<string, []>("op_3199_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3199_cast = slice_by_index(begin = var_3199_begin_0, end = var_3199_end_0, end_mask = var_3199_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3199_cast")];
            tensor<int32, [4]> var_3203_begin_0 = const()[name = tensor<string, []>("op_3203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_3203_end_0 = const()[name = tensor<string, []>("op_3203_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_3203_end_mask_0 = const()[name = tensor<string, []>("op_3203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3203_cast = slice_by_index(begin = var_3203_begin_0, end = var_3203_end_0, end_mask = var_3203_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3203_cast")];
            tensor<int32, [4]> var_3207_begin_0 = const()[name = tensor<string, []>("op_3207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_3207_end_0 = const()[name = tensor<string, []>("op_3207_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_3207_end_mask_0 = const()[name = tensor<string, []>("op_3207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3207_cast = slice_by_index(begin = var_3207_begin_0, end = var_3207_end_0, end_mask = var_3207_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3207_cast")];
            tensor<int32, [4]> var_3211_begin_0 = const()[name = tensor<string, []>("op_3211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_3211_end_0 = const()[name = tensor<string, []>("op_3211_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_3211_end_mask_0 = const()[name = tensor<string, []>("op_3211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3211_cast = slice_by_index(begin = var_3211_begin_0, end = var_3211_end_0, end_mask = var_3211_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3211_cast")];
            tensor<int32, [4]> var_3215_begin_0 = const()[name = tensor<string, []>("op_3215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_3215_end_0 = const()[name = tensor<string, []>("op_3215_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_3215_end_mask_0 = const()[name = tensor<string, []>("op_3215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3215_cast = slice_by_index(begin = var_3215_begin_0, end = var_3215_end_0, end_mask = var_3215_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3215_cast")];
            tensor<int32, [4]> var_3219_begin_0 = const()[name = tensor<string, []>("op_3219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3219_end_0 = const()[name = tensor<string, []>("op_3219_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_3219_end_mask_0 = const()[name = tensor<string, []>("op_3219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3219_cast = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3219_cast")];
            tensor<int32, [4]> var_3223_begin_0 = const()[name = tensor<string, []>("op_3223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_3223_end_0 = const()[name = tensor<string, []>("op_3223_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_3223_end_mask_0 = const()[name = tensor<string, []>("op_3223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_3223_cast = slice_by_index(begin = var_3223_begin_0, end = var_3223_end_0, end_mask = var_3223_end_mask_0, x = transpose_25)[name = tensor<string, []>("op_3223_cast")];
            tensor<int32, [4]> var_3225_begin_0 = const()[name = tensor<string, []>("op_3225_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3225_end_0 = const()[name = tensor<string, []>("op_3225_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3225_end_mask_0 = const()[name = tensor<string, []>("op_3225_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3225_cast = slice_by_index(begin = var_3225_begin_0, end = var_3225_end_0, end_mask = var_3225_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3225_cast")];
            tensor<int32, [4]> var_3229_begin_0 = const()[name = tensor<string, []>("op_3229_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3229_end_0 = const()[name = tensor<string, []>("op_3229_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_3229_end_mask_0 = const()[name = tensor<string, []>("op_3229_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3229_cast = slice_by_index(begin = var_3229_begin_0, end = var_3229_end_0, end_mask = var_3229_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3229_cast")];
            tensor<int32, [4]> var_3233_begin_0 = const()[name = tensor<string, []>("op_3233_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3233_end_0 = const()[name = tensor<string, []>("op_3233_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_3233_end_mask_0 = const()[name = tensor<string, []>("op_3233_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3233_cast = slice_by_index(begin = var_3233_begin_0, end = var_3233_end_0, end_mask = var_3233_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3233_cast")];
            tensor<int32, [4]> var_3237_begin_0 = const()[name = tensor<string, []>("op_3237_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3237_end_0 = const()[name = tensor<string, []>("op_3237_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3237_end_mask_0 = const()[name = tensor<string, []>("op_3237_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3237_cast = slice_by_index(begin = var_3237_begin_0, end = var_3237_end_0, end_mask = var_3237_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3237_cast")];
            tensor<int32, [4]> var_3241_begin_0 = const()[name = tensor<string, []>("op_3241_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3241_end_0 = const()[name = tensor<string, []>("op_3241_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3241_end_mask_0 = const()[name = tensor<string, []>("op_3241_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3241_cast = slice_by_index(begin = var_3241_begin_0, end = var_3241_end_0, end_mask = var_3241_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3241_cast")];
            tensor<int32, [4]> var_3245_begin_0 = const()[name = tensor<string, []>("op_3245_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3245_end_0 = const()[name = tensor<string, []>("op_3245_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_3245_end_mask_0 = const()[name = tensor<string, []>("op_3245_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3245_cast = slice_by_index(begin = var_3245_begin_0, end = var_3245_end_0, end_mask = var_3245_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3245_cast")];
            tensor<int32, [4]> var_3249_begin_0 = const()[name = tensor<string, []>("op_3249_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3249_end_0 = const()[name = tensor<string, []>("op_3249_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_3249_end_mask_0 = const()[name = tensor<string, []>("op_3249_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3249_cast = slice_by_index(begin = var_3249_begin_0, end = var_3249_end_0, end_mask = var_3249_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3249_cast")];
            tensor<int32, [4]> var_3253_begin_0 = const()[name = tensor<string, []>("op_3253_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3253_end_0 = const()[name = tensor<string, []>("op_3253_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_3253_end_mask_0 = const()[name = tensor<string, []>("op_3253_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3253_cast = slice_by_index(begin = var_3253_begin_0, end = var_3253_end_0, end_mask = var_3253_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3253_cast")];
            tensor<int32, [4]> var_3257_begin_0 = const()[name = tensor<string, []>("op_3257_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3257_end_0 = const()[name = tensor<string, []>("op_3257_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_3257_end_mask_0 = const()[name = tensor<string, []>("op_3257_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3257_cast = slice_by_index(begin = var_3257_begin_0, end = var_3257_end_0, end_mask = var_3257_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3257_cast")];
            tensor<int32, [4]> var_3261_begin_0 = const()[name = tensor<string, []>("op_3261_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3261_end_0 = const()[name = tensor<string, []>("op_3261_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3261_end_mask_0 = const()[name = tensor<string, []>("op_3261_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3261_cast = slice_by_index(begin = var_3261_begin_0, end = var_3261_end_0, end_mask = var_3261_end_mask_0, x = v_13_cast)[name = tensor<string, []>("op_3261_cast")];
            tensor<string, []> var_3265_equation_0 = const()[name = tensor<string, []>("op_3265_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3265_cast = einsum(equation = var_3265_equation_0, values = (var_3187_cast, var_3163_cast))[name = tensor<string, []>("op_3265_cast")];
            tensor<fp16, []> var_3266_to_fp16 = const()[name = tensor<string, []>("op_3266_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_401_cast = mul(x = var_3265_cast, y = var_3266_to_fp16)[name = tensor<string, []>("aw_chunk_401_cast")];
            tensor<string, []> var_3269_equation_0 = const()[name = tensor<string, []>("op_3269_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3269_cast = einsum(equation = var_3269_equation_0, values = (var_3187_cast, var_3164_cast))[name = tensor<string, []>("op_3269_cast")];
            tensor<fp16, []> var_3270_to_fp16 = const()[name = tensor<string, []>("op_3270_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_403_cast = mul(x = var_3269_cast, y = var_3270_to_fp16)[name = tensor<string, []>("aw_chunk_403_cast")];
            tensor<string, []> var_3273_equation_0 = const()[name = tensor<string, []>("op_3273_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3273_cast = einsum(equation = var_3273_equation_0, values = (var_3191_cast, var_3165_cast))[name = tensor<string, []>("op_3273_cast")];
            tensor<fp16, []> var_3274_to_fp16 = const()[name = tensor<string, []>("op_3274_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_405_cast = mul(x = var_3273_cast, y = var_3274_to_fp16)[name = tensor<string, []>("aw_chunk_405_cast")];
            tensor<string, []> var_3277_equation_0 = const()[name = tensor<string, []>("op_3277_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3277_cast = einsum(equation = var_3277_equation_0, values = (var_3191_cast, var_3166_cast))[name = tensor<string, []>("op_3277_cast")];
            tensor<fp16, []> var_3278_to_fp16 = const()[name = tensor<string, []>("op_3278_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_407_cast = mul(x = var_3277_cast, y = var_3278_to_fp16)[name = tensor<string, []>("aw_chunk_407_cast")];
            tensor<string, []> var_3281_equation_0 = const()[name = tensor<string, []>("op_3281_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3281_cast = einsum(equation = var_3281_equation_0, values = (var_3195_cast, var_3167_cast))[name = tensor<string, []>("op_3281_cast")];
            tensor<fp16, []> var_3282_to_fp16 = const()[name = tensor<string, []>("op_3282_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_409_cast = mul(x = var_3281_cast, y = var_3282_to_fp16)[name = tensor<string, []>("aw_chunk_409_cast")];
            tensor<string, []> var_3285_equation_0 = const()[name = tensor<string, []>("op_3285_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3285_cast = einsum(equation = var_3285_equation_0, values = (var_3195_cast, var_3168_cast))[name = tensor<string, []>("op_3285_cast")];
            tensor<fp16, []> var_3286_to_fp16 = const()[name = tensor<string, []>("op_3286_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_411_cast = mul(x = var_3285_cast, y = var_3286_to_fp16)[name = tensor<string, []>("aw_chunk_411_cast")];
            tensor<string, []> var_3289_equation_0 = const()[name = tensor<string, []>("op_3289_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3289_cast = einsum(equation = var_3289_equation_0, values = (var_3199_cast, var_3169_cast))[name = tensor<string, []>("op_3289_cast")];
            tensor<fp16, []> var_3290_to_fp16 = const()[name = tensor<string, []>("op_3290_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_413_cast = mul(x = var_3289_cast, y = var_3290_to_fp16)[name = tensor<string, []>("aw_chunk_413_cast")];
            tensor<string, []> var_3293_equation_0 = const()[name = tensor<string, []>("op_3293_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3293_cast = einsum(equation = var_3293_equation_0, values = (var_3199_cast, var_3170_cast))[name = tensor<string, []>("op_3293_cast")];
            tensor<fp16, []> var_3294_to_fp16 = const()[name = tensor<string, []>("op_3294_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_415_cast = mul(x = var_3293_cast, y = var_3294_to_fp16)[name = tensor<string, []>("aw_chunk_415_cast")];
            tensor<string, []> var_3297_equation_0 = const()[name = tensor<string, []>("op_3297_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3297_cast = einsum(equation = var_3297_equation_0, values = (var_3203_cast, var_3171_cast))[name = tensor<string, []>("op_3297_cast")];
            tensor<fp16, []> var_3298_to_fp16 = const()[name = tensor<string, []>("op_3298_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_417_cast = mul(x = var_3297_cast, y = var_3298_to_fp16)[name = tensor<string, []>("aw_chunk_417_cast")];
            tensor<string, []> var_3301_equation_0 = const()[name = tensor<string, []>("op_3301_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3301_cast = einsum(equation = var_3301_equation_0, values = (var_3203_cast, var_3172_cast))[name = tensor<string, []>("op_3301_cast")];
            tensor<fp16, []> var_3302_to_fp16 = const()[name = tensor<string, []>("op_3302_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_419_cast = mul(x = var_3301_cast, y = var_3302_to_fp16)[name = tensor<string, []>("aw_chunk_419_cast")];
            tensor<string, []> var_3305_equation_0 = const()[name = tensor<string, []>("op_3305_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3305_cast = einsum(equation = var_3305_equation_0, values = (var_3207_cast, var_3173_cast))[name = tensor<string, []>("op_3305_cast")];
            tensor<fp16, []> var_3306_to_fp16 = const()[name = tensor<string, []>("op_3306_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_421_cast = mul(x = var_3305_cast, y = var_3306_to_fp16)[name = tensor<string, []>("aw_chunk_421_cast")];
            tensor<string, []> var_3309_equation_0 = const()[name = tensor<string, []>("op_3309_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3309_cast = einsum(equation = var_3309_equation_0, values = (var_3207_cast, var_3174_cast))[name = tensor<string, []>("op_3309_cast")];
            tensor<fp16, []> var_3310_to_fp16 = const()[name = tensor<string, []>("op_3310_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_423_cast = mul(x = var_3309_cast, y = var_3310_to_fp16)[name = tensor<string, []>("aw_chunk_423_cast")];
            tensor<string, []> var_3313_equation_0 = const()[name = tensor<string, []>("op_3313_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3313_cast = einsum(equation = var_3313_equation_0, values = (var_3211_cast, var_3175_cast))[name = tensor<string, []>("op_3313_cast")];
            tensor<fp16, []> var_3314_to_fp16 = const()[name = tensor<string, []>("op_3314_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_425_cast = mul(x = var_3313_cast, y = var_3314_to_fp16)[name = tensor<string, []>("aw_chunk_425_cast")];
            tensor<string, []> var_3317_equation_0 = const()[name = tensor<string, []>("op_3317_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3317_cast = einsum(equation = var_3317_equation_0, values = (var_3211_cast, var_3176_cast))[name = tensor<string, []>("op_3317_cast")];
            tensor<fp16, []> var_3318_to_fp16 = const()[name = tensor<string, []>("op_3318_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_427_cast = mul(x = var_3317_cast, y = var_3318_to_fp16)[name = tensor<string, []>("aw_chunk_427_cast")];
            tensor<string, []> var_3321_equation_0 = const()[name = tensor<string, []>("op_3321_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3321_cast = einsum(equation = var_3321_equation_0, values = (var_3215_cast, var_3177_cast))[name = tensor<string, []>("op_3321_cast")];
            tensor<fp16, []> var_3322_to_fp16 = const()[name = tensor<string, []>("op_3322_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_429_cast = mul(x = var_3321_cast, y = var_3322_to_fp16)[name = tensor<string, []>("aw_chunk_429_cast")];
            tensor<string, []> var_3325_equation_0 = const()[name = tensor<string, []>("op_3325_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3325_cast = einsum(equation = var_3325_equation_0, values = (var_3215_cast, var_3178_cast))[name = tensor<string, []>("op_3325_cast")];
            tensor<fp16, []> var_3326_to_fp16 = const()[name = tensor<string, []>("op_3326_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_431_cast = mul(x = var_3325_cast, y = var_3326_to_fp16)[name = tensor<string, []>("aw_chunk_431_cast")];
            tensor<string, []> var_3329_equation_0 = const()[name = tensor<string, []>("op_3329_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3329_cast = einsum(equation = var_3329_equation_0, values = (var_3219_cast, var_3179_cast))[name = tensor<string, []>("op_3329_cast")];
            tensor<fp16, []> var_3330_to_fp16 = const()[name = tensor<string, []>("op_3330_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_433_cast = mul(x = var_3329_cast, y = var_3330_to_fp16)[name = tensor<string, []>("aw_chunk_433_cast")];
            tensor<string, []> var_3333_equation_0 = const()[name = tensor<string, []>("op_3333_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3333_cast = einsum(equation = var_3333_equation_0, values = (var_3219_cast, var_3180_cast))[name = tensor<string, []>("op_3333_cast")];
            tensor<fp16, []> var_3334_to_fp16 = const()[name = tensor<string, []>("op_3334_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_435_cast = mul(x = var_3333_cast, y = var_3334_to_fp16)[name = tensor<string, []>("aw_chunk_435_cast")];
            tensor<string, []> var_3337_equation_0 = const()[name = tensor<string, []>("op_3337_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3337_cast = einsum(equation = var_3337_equation_0, values = (var_3223_cast, var_3181_cast))[name = tensor<string, []>("op_3337_cast")];
            tensor<fp16, []> var_3338_to_fp16 = const()[name = tensor<string, []>("op_3338_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_437_cast = mul(x = var_3337_cast, y = var_3338_to_fp16)[name = tensor<string, []>("aw_chunk_437_cast")];
            tensor<string, []> var_3341_equation_0 = const()[name = tensor<string, []>("op_3341_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_3341_cast = einsum(equation = var_3341_equation_0, values = (var_3223_cast, var_3182_cast))[name = tensor<string, []>("op_3341_cast")];
            tensor<fp16, []> var_3342_to_fp16 = const()[name = tensor<string, []>("op_3342_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_439_cast = mul(x = var_3341_cast, y = var_3342_to_fp16)[name = tensor<string, []>("aw_chunk_439_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3344_cast = softmax(axis = var_2173, x = aw_chunk_401_cast)[name = tensor<string, []>("op_3344_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3345_cast = softmax(axis = var_2173, x = aw_chunk_403_cast)[name = tensor<string, []>("op_3345_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3346_cast = softmax(axis = var_2173, x = aw_chunk_405_cast)[name = tensor<string, []>("op_3346_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3347_cast = softmax(axis = var_2173, x = aw_chunk_407_cast)[name = tensor<string, []>("op_3347_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3348_cast = softmax(axis = var_2173, x = aw_chunk_409_cast)[name = tensor<string, []>("op_3348_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3349_cast = softmax(axis = var_2173, x = aw_chunk_411_cast)[name = tensor<string, []>("op_3349_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3350_cast = softmax(axis = var_2173, x = aw_chunk_413_cast)[name = tensor<string, []>("op_3350_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3351_cast = softmax(axis = var_2173, x = aw_chunk_415_cast)[name = tensor<string, []>("op_3351_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3352_cast = softmax(axis = var_2173, x = aw_chunk_417_cast)[name = tensor<string, []>("op_3352_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3353_cast = softmax(axis = var_2173, x = aw_chunk_419_cast)[name = tensor<string, []>("op_3353_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3354_cast = softmax(axis = var_2173, x = aw_chunk_421_cast)[name = tensor<string, []>("op_3354_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3355_cast = softmax(axis = var_2173, x = aw_chunk_423_cast)[name = tensor<string, []>("op_3355_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3356_cast = softmax(axis = var_2173, x = aw_chunk_425_cast)[name = tensor<string, []>("op_3356_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3357_cast = softmax(axis = var_2173, x = aw_chunk_427_cast)[name = tensor<string, []>("op_3357_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3358_cast = softmax(axis = var_2173, x = aw_chunk_429_cast)[name = tensor<string, []>("op_3358_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3359_cast = softmax(axis = var_2173, x = aw_chunk_431_cast)[name = tensor<string, []>("op_3359_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3360_cast = softmax(axis = var_2173, x = aw_chunk_433_cast)[name = tensor<string, []>("op_3360_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3361_cast = softmax(axis = var_2173, x = aw_chunk_435_cast)[name = tensor<string, []>("op_3361_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3362_cast = softmax(axis = var_2173, x = aw_chunk_437_cast)[name = tensor<string, []>("op_3362_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_3363_cast = softmax(axis = var_2173, x = aw_chunk_439_cast)[name = tensor<string, []>("op_3363_cast")];
            tensor<string, []> var_3365_equation_0 = const()[name = tensor<string, []>("op_3365_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3365_cast = einsum(equation = var_3365_equation_0, values = (var_3225_cast, var_3344_cast))[name = tensor<string, []>("op_3365_cast")];
            tensor<string, []> var_3367_equation_0 = const()[name = tensor<string, []>("op_3367_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3367_cast = einsum(equation = var_3367_equation_0, values = (var_3225_cast, var_3345_cast))[name = tensor<string, []>("op_3367_cast")];
            tensor<string, []> var_3369_equation_0 = const()[name = tensor<string, []>("op_3369_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3369_cast = einsum(equation = var_3369_equation_0, values = (var_3229_cast, var_3346_cast))[name = tensor<string, []>("op_3369_cast")];
            tensor<string, []> var_3371_equation_0 = const()[name = tensor<string, []>("op_3371_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3371_cast = einsum(equation = var_3371_equation_0, values = (var_3229_cast, var_3347_cast))[name = tensor<string, []>("op_3371_cast")];
            tensor<string, []> var_3373_equation_0 = const()[name = tensor<string, []>("op_3373_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3373_cast = einsum(equation = var_3373_equation_0, values = (var_3233_cast, var_3348_cast))[name = tensor<string, []>("op_3373_cast")];
            tensor<string, []> var_3375_equation_0 = const()[name = tensor<string, []>("op_3375_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3375_cast = einsum(equation = var_3375_equation_0, values = (var_3233_cast, var_3349_cast))[name = tensor<string, []>("op_3375_cast")];
            tensor<string, []> var_3377_equation_0 = const()[name = tensor<string, []>("op_3377_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3377_cast = einsum(equation = var_3377_equation_0, values = (var_3237_cast, var_3350_cast))[name = tensor<string, []>("op_3377_cast")];
            tensor<string, []> var_3379_equation_0 = const()[name = tensor<string, []>("op_3379_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3379_cast = einsum(equation = var_3379_equation_0, values = (var_3237_cast, var_3351_cast))[name = tensor<string, []>("op_3379_cast")];
            tensor<string, []> var_3381_equation_0 = const()[name = tensor<string, []>("op_3381_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3381_cast = einsum(equation = var_3381_equation_0, values = (var_3241_cast, var_3352_cast))[name = tensor<string, []>("op_3381_cast")];
            tensor<string, []> var_3383_equation_0 = const()[name = tensor<string, []>("op_3383_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3383_cast = einsum(equation = var_3383_equation_0, values = (var_3241_cast, var_3353_cast))[name = tensor<string, []>("op_3383_cast")];
            tensor<string, []> var_3385_equation_0 = const()[name = tensor<string, []>("op_3385_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3385_cast = einsum(equation = var_3385_equation_0, values = (var_3245_cast, var_3354_cast))[name = tensor<string, []>("op_3385_cast")];
            tensor<string, []> var_3387_equation_0 = const()[name = tensor<string, []>("op_3387_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3387_cast = einsum(equation = var_3387_equation_0, values = (var_3245_cast, var_3355_cast))[name = tensor<string, []>("op_3387_cast")];
            tensor<string, []> var_3389_equation_0 = const()[name = tensor<string, []>("op_3389_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3389_cast = einsum(equation = var_3389_equation_0, values = (var_3249_cast, var_3356_cast))[name = tensor<string, []>("op_3389_cast")];
            tensor<string, []> var_3391_equation_0 = const()[name = tensor<string, []>("op_3391_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3391_cast = einsum(equation = var_3391_equation_0, values = (var_3249_cast, var_3357_cast))[name = tensor<string, []>("op_3391_cast")];
            tensor<string, []> var_3393_equation_0 = const()[name = tensor<string, []>("op_3393_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3393_cast = einsum(equation = var_3393_equation_0, values = (var_3253_cast, var_3358_cast))[name = tensor<string, []>("op_3393_cast")];
            tensor<string, []> var_3395_equation_0 = const()[name = tensor<string, []>("op_3395_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3395_cast = einsum(equation = var_3395_equation_0, values = (var_3253_cast, var_3359_cast))[name = tensor<string, []>("op_3395_cast")];
            tensor<string, []> var_3397_equation_0 = const()[name = tensor<string, []>("op_3397_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3397_cast = einsum(equation = var_3397_equation_0, values = (var_3257_cast, var_3360_cast))[name = tensor<string, []>("op_3397_cast")];
            tensor<string, []> var_3399_equation_0 = const()[name = tensor<string, []>("op_3399_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3399_cast = einsum(equation = var_3399_equation_0, values = (var_3257_cast, var_3361_cast))[name = tensor<string, []>("op_3399_cast")];
            tensor<string, []> var_3401_equation_0 = const()[name = tensor<string, []>("op_3401_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3401_cast = einsum(equation = var_3401_equation_0, values = (var_3261_cast, var_3362_cast))[name = tensor<string, []>("op_3401_cast")];
            tensor<string, []> var_3403_equation_0 = const()[name = tensor<string, []>("op_3403_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3403_cast = einsum(equation = var_3403_equation_0, values = (var_3261_cast, var_3363_cast))[name = tensor<string, []>("op_3403_cast")];
            tensor<bool, []> var_3405_interleave_0 = const()[name = tensor<string, []>("op_3405_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3405_cast = concat(axis = var_2150, interleave = var_3405_interleave_0, values = (var_3365_cast, var_3367_cast))[name = tensor<string, []>("op_3405_cast")];
            tensor<bool, []> var_3407_interleave_0 = const()[name = tensor<string, []>("op_3407_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3407_cast = concat(axis = var_2150, interleave = var_3407_interleave_0, values = (var_3369_cast, var_3371_cast))[name = tensor<string, []>("op_3407_cast")];
            tensor<bool, []> var_3409_interleave_0 = const()[name = tensor<string, []>("op_3409_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3409_cast = concat(axis = var_2150, interleave = var_3409_interleave_0, values = (var_3373_cast, var_3375_cast))[name = tensor<string, []>("op_3409_cast")];
            tensor<bool, []> var_3411_interleave_0 = const()[name = tensor<string, []>("op_3411_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3411_cast = concat(axis = var_2150, interleave = var_3411_interleave_0, values = (var_3377_cast, var_3379_cast))[name = tensor<string, []>("op_3411_cast")];
            tensor<bool, []> var_3413_interleave_0 = const()[name = tensor<string, []>("op_3413_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3413_cast = concat(axis = var_2150, interleave = var_3413_interleave_0, values = (var_3381_cast, var_3383_cast))[name = tensor<string, []>("op_3413_cast")];
            tensor<bool, []> var_3415_interleave_0 = const()[name = tensor<string, []>("op_3415_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3415_cast = concat(axis = var_2150, interleave = var_3415_interleave_0, values = (var_3385_cast, var_3387_cast))[name = tensor<string, []>("op_3415_cast")];
            tensor<bool, []> var_3417_interleave_0 = const()[name = tensor<string, []>("op_3417_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3417_cast = concat(axis = var_2150, interleave = var_3417_interleave_0, values = (var_3389_cast, var_3391_cast))[name = tensor<string, []>("op_3417_cast")];
            tensor<bool, []> var_3419_interleave_0 = const()[name = tensor<string, []>("op_3419_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3419_cast = concat(axis = var_2150, interleave = var_3419_interleave_0, values = (var_3393_cast, var_3395_cast))[name = tensor<string, []>("op_3419_cast")];
            tensor<bool, []> var_3421_interleave_0 = const()[name = tensor<string, []>("op_3421_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3421_cast = concat(axis = var_2150, interleave = var_3421_interleave_0, values = (var_3397_cast, var_3399_cast))[name = tensor<string, []>("op_3421_cast")];
            tensor<bool, []> var_3423_interleave_0 = const()[name = tensor<string, []>("op_3423_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3423_cast = concat(axis = var_2150, interleave = var_3423_interleave_0, values = (var_3401_cast, var_3403_cast))[name = tensor<string, []>("op_3423_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_2173, interleave = input_105_interleave_0, values = (var_3405_cast, var_3407_cast, var_3409_cast, var_3411_cast, var_3413_cast, var_3415_cast, var_3417_cast, var_3419_cast, var_3421_cast, var_3423_cast))[name = tensor<string, []>("input_105_cast")];
            tensor<int32, [2]> var_3429 = const()[name = tensor<string, []>("op_3429"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3431 = const()[name = tensor<string, []>("op_3431"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3433_pad_type_0 = const()[name = tensor<string, []>("op_3433_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3433_pad_0 = const()[name = tensor<string, []>("op_3433_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9783872))), lut = tensor<fp16, [4]>([-0x1.568p-4, -0x1.99cp-6, 0x1.918p-6, 0x1.544p-4]), 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, []>(9886336)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3433_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_3431, groups = var_2173, pad = var_3433_pad_0, pad_type = var_3433_pad_type_0, strides = var_3429, 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_3433_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_21_cast = add(x = var_3433_cast, y = inputs_19_cast)[name = tensor<string, []>("inputs_21_cast")];
            tensor<int32, [1]> var_3437 = const()[name = tensor<string, []>("op_3437"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_21_cast = reduce_mean(axes = var_3437, keep_dims = var_2168, 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_3441 = const()[name = tensor<string, []>("op_3441"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3442_cast = reduce_mean(axes = var_3441, keep_dims = var_2168, x = zero_mean_sq_21_cast)[name = tensor<string, []>("op_3442_cast")];
            tensor<fp16, []> var_3443_to_fp16 = const()[name = tensor<string, []>("op_3443_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3444_cast = add(x = var_3442_cast, y = var_3443_to_fp16)[name = tensor<string, []>("op_3444_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_3444_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_3448_to_fp16 = const()[name = tensor<string, []>("op_3448_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9887680)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3449_cast = add(x = out_21_cast, y = var_3448_to_fp16)[name = tensor<string, []>("op_3449_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3451_to_fp16 = const()[name = tensor<string, []>("op_3451_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9889024)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_65_cast = mul(x = var_3449_cast, y = var_3451_to_fp16)[name = tensor<string, []>("hidden_states_65_cast")];
            tensor<int32, [2]> var_3458 = const()[name = tensor<string, []>("op_3458"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3460 = const()[name = tensor<string, []>("op_3460"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9890368))), lut = tensor<fp16, [4]>([-0x1.724p-4, -0x1.ba8p-6, 0x1.bbcp-6, 0x1.72cp-4]), 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_3460, groups = var_2173, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = var_3458, 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_3464 = const()[name = tensor<string, []>("op_3464"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3466 = const()[name = tensor<string, []>("op_3466"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_29_pad_type_0 = const()[name = tensor<string, []>("k_29_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_29_pad_0 = const()[name = tensor<string, []>("k_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9992832))), lut = tensor<fp16, [4]>([-0x1.5p-4, -0x1.8d4p-6, 0x1.8ecp-6, 0x1.51p-4]), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_29_cast = conv(dilations = var_3466, groups = var_2173, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = var_3464, 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_3470 = const()[name = tensor<string, []>("op_3470"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3472 = const()[name = tensor<string, []>("op_3472"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_15_pad_type_0 = const()[name = tensor<string, []>("v_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_15_pad_0 = const()[name = tensor<string, []>("v_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10156736))), lut = tensor<fp16, [4]>([-0x1.414p-5, -0x1.6d8p-7, 0x1.6d4p-7, 0x1.41cp-5]), name = tensor<string, []>("down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_15_cast = conv(dilations = var_3472, groups = var_2173, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = var_3470, 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_3476_begin_0 = const()[name = tensor<string, []>("op_3476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3476_end_0 = const()[name = tensor<string, []>("op_3476_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3476_end_mask_0 = const()[name = tensor<string, []>("op_3476_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3476_cast = slice_by_index(begin = var_3476_begin_0, end = var_3476_end_0, end_mask = var_3476_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3476_cast")];
            tensor<int32, [4]> var_3480_begin_0 = const()[name = tensor<string, []>("op_3480_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3480_end_0 = const()[name = tensor<string, []>("op_3480_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_3480_end_mask_0 = const()[name = tensor<string, []>("op_3480_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3480_cast = slice_by_index(begin = var_3480_begin_0, end = var_3480_end_0, end_mask = var_3480_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3480_cast")];
            tensor<int32, [4]> var_3484_begin_0 = const()[name = tensor<string, []>("op_3484_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3484_end_0 = const()[name = tensor<string, []>("op_3484_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_3484_end_mask_0 = const()[name = tensor<string, []>("op_3484_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3484_cast = slice_by_index(begin = var_3484_begin_0, end = var_3484_end_0, end_mask = var_3484_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3484_cast")];
            tensor<int32, [4]> var_3488_begin_0 = const()[name = tensor<string, []>("op_3488_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3488_end_0 = const()[name = tensor<string, []>("op_3488_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_3488_end_mask_0 = const()[name = tensor<string, []>("op_3488_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3488_cast = slice_by_index(begin = var_3488_begin_0, end = var_3488_end_0, end_mask = var_3488_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3488_cast")];
            tensor<int32, [4]> var_3492_begin_0 = const()[name = tensor<string, []>("op_3492_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3492_end_0 = const()[name = tensor<string, []>("op_3492_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_3492_end_mask_0 = const()[name = tensor<string, []>("op_3492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3492_cast = slice_by_index(begin = var_3492_begin_0, end = var_3492_end_0, end_mask = var_3492_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3492_cast")];
            tensor<int32, [4]> var_3496_begin_0 = const()[name = tensor<string, []>("op_3496_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3496_end_0 = const()[name = tensor<string, []>("op_3496_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_3496_end_mask_0 = const()[name = tensor<string, []>("op_3496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3496_cast = slice_by_index(begin = var_3496_begin_0, end = var_3496_end_0, end_mask = var_3496_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3496_cast")];
            tensor<int32, [4]> var_3500_begin_0 = const()[name = tensor<string, []>("op_3500_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3500_end_0 = const()[name = tensor<string, []>("op_3500_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_3500_end_mask_0 = const()[name = tensor<string, []>("op_3500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3500_cast = slice_by_index(begin = var_3500_begin_0, end = var_3500_end_0, end_mask = var_3500_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3500_cast")];
            tensor<int32, [4]> var_3504_begin_0 = const()[name = tensor<string, []>("op_3504_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3504_end_0 = const()[name = tensor<string, []>("op_3504_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_3504_end_mask_0 = const()[name = tensor<string, []>("op_3504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3504_cast = slice_by_index(begin = var_3504_begin_0, end = var_3504_end_0, end_mask = var_3504_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3504_cast")];
            tensor<int32, [4]> var_3508_begin_0 = const()[name = tensor<string, []>("op_3508_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3508_end_0 = const()[name = tensor<string, []>("op_3508_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_3508_end_mask_0 = const()[name = tensor<string, []>("op_3508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3508_cast = slice_by_index(begin = var_3508_begin_0, end = var_3508_end_0, end_mask = var_3508_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3508_cast")];
            tensor<int32, [4]> var_3512_begin_0 = const()[name = tensor<string, []>("op_3512_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3512_end_0 = const()[name = tensor<string, []>("op_3512_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_3512_end_mask_0 = const()[name = tensor<string, []>("op_3512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_3512_cast = slice_by_index(begin = var_3512_begin_0, end = var_3512_end_0, end_mask = var_3512_end_mask_0, x = q_15_cast)[name = tensor<string, []>("op_3512_cast")];
            tensor<int32, [4]> var_3515_begin_0 = const()[name = tensor<string, []>("op_3515_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3515_end_0 = const()[name = tensor<string, []>("op_3515_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3515_end_mask_0 = const()[name = tensor<string, []>("op_3515_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3515_cast = slice_by_index(begin = var_3515_begin_0, end = var_3515_end_0, end_mask = var_3515_end_mask_0, x = var_3476_cast)[name = tensor<string, []>("op_3515_cast")];
            tensor<int32, [4]> var_3516_begin_0 = const()[name = tensor<string, []>("op_3516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3516_end_0 = const()[name = tensor<string, []>("op_3516_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3516_end_mask_0 = const()[name = tensor<string, []>("op_3516_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3516_cast = slice_by_index(begin = var_3516_begin_0, end = var_3516_end_0, end_mask = var_3516_end_mask_0, x = var_3476_cast)[name = tensor<string, []>("op_3516_cast")];
            tensor<int32, [4]> var_3517_begin_0 = const()[name = tensor<string, []>("op_3517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3517_end_0 = const()[name = tensor<string, []>("op_3517_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3517_end_mask_0 = const()[name = tensor<string, []>("op_3517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3517_cast = slice_by_index(begin = var_3517_begin_0, end = var_3517_end_0, end_mask = var_3517_end_mask_0, x = var_3480_cast)[name = tensor<string, []>("op_3517_cast")];
            tensor<int32, [4]> var_3518_begin_0 = const()[name = tensor<string, []>("op_3518_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3518_end_0 = const()[name = tensor<string, []>("op_3518_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3518_end_mask_0 = const()[name = tensor<string, []>("op_3518_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3518_cast = slice_by_index(begin = var_3518_begin_0, end = var_3518_end_0, end_mask = var_3518_end_mask_0, x = var_3480_cast)[name = tensor<string, []>("op_3518_cast")];
            tensor<int32, [4]> var_3519_begin_0 = const()[name = tensor<string, []>("op_3519_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3519_end_0 = const()[name = tensor<string, []>("op_3519_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3519_end_mask_0 = const()[name = tensor<string, []>("op_3519_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3519_cast = slice_by_index(begin = var_3519_begin_0, end = var_3519_end_0, end_mask = var_3519_end_mask_0, x = var_3484_cast)[name = tensor<string, []>("op_3519_cast")];
            tensor<int32, [4]> var_3520_begin_0 = const()[name = tensor<string, []>("op_3520_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3520_end_0 = const()[name = tensor<string, []>("op_3520_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3520_end_mask_0 = const()[name = tensor<string, []>("op_3520_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3520_cast = slice_by_index(begin = var_3520_begin_0, end = var_3520_end_0, end_mask = var_3520_end_mask_0, x = var_3484_cast)[name = tensor<string, []>("op_3520_cast")];
            tensor<int32, [4]> var_3521_begin_0 = const()[name = tensor<string, []>("op_3521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3521_end_0 = const()[name = tensor<string, []>("op_3521_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3521_end_mask_0 = const()[name = tensor<string, []>("op_3521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3521_cast = slice_by_index(begin = var_3521_begin_0, end = var_3521_end_0, end_mask = var_3521_end_mask_0, x = var_3488_cast)[name = tensor<string, []>("op_3521_cast")];
            tensor<int32, [4]> var_3522_begin_0 = const()[name = tensor<string, []>("op_3522_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3522_end_0 = const()[name = tensor<string, []>("op_3522_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3522_end_mask_0 = const()[name = tensor<string, []>("op_3522_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3522_cast = slice_by_index(begin = var_3522_begin_0, end = var_3522_end_0, end_mask = var_3522_end_mask_0, x = var_3488_cast)[name = tensor<string, []>("op_3522_cast")];
            tensor<int32, [4]> var_3523_begin_0 = const()[name = tensor<string, []>("op_3523_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3523_end_0 = const()[name = tensor<string, []>("op_3523_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3523_end_mask_0 = const()[name = tensor<string, []>("op_3523_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3523_cast = slice_by_index(begin = var_3523_begin_0, end = var_3523_end_0, end_mask = var_3523_end_mask_0, x = var_3492_cast)[name = tensor<string, []>("op_3523_cast")];
            tensor<int32, [4]> var_3524_begin_0 = const()[name = tensor<string, []>("op_3524_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3524_end_0 = const()[name = tensor<string, []>("op_3524_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3524_end_mask_0 = const()[name = tensor<string, []>("op_3524_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3524_cast = slice_by_index(begin = var_3524_begin_0, end = var_3524_end_0, end_mask = var_3524_end_mask_0, x = var_3492_cast)[name = tensor<string, []>("op_3524_cast")];
            tensor<int32, [4]> var_3525_begin_0 = const()[name = tensor<string, []>("op_3525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3525_end_0 = const()[name = tensor<string, []>("op_3525_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3525_end_mask_0 = const()[name = tensor<string, []>("op_3525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3525_cast = slice_by_index(begin = var_3525_begin_0, end = var_3525_end_0, end_mask = var_3525_end_mask_0, x = var_3496_cast)[name = tensor<string, []>("op_3525_cast")];
            tensor<int32, [4]> var_3526_begin_0 = const()[name = tensor<string, []>("op_3526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3526_end_0 = const()[name = tensor<string, []>("op_3526_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3526_end_mask_0 = const()[name = tensor<string, []>("op_3526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3526_cast = slice_by_index(begin = var_3526_begin_0, end = var_3526_end_0, end_mask = var_3526_end_mask_0, x = var_3496_cast)[name = tensor<string, []>("op_3526_cast")];
            tensor<int32, [4]> var_3527_begin_0 = const()[name = tensor<string, []>("op_3527_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3527_end_0 = const()[name = tensor<string, []>("op_3527_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3527_end_mask_0 = const()[name = tensor<string, []>("op_3527_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3527_cast = slice_by_index(begin = var_3527_begin_0, end = var_3527_end_0, end_mask = var_3527_end_mask_0, x = var_3500_cast)[name = tensor<string, []>("op_3527_cast")];
            tensor<int32, [4]> var_3528_begin_0 = const()[name = tensor<string, []>("op_3528_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3528_end_0 = const()[name = tensor<string, []>("op_3528_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3528_end_mask_0 = const()[name = tensor<string, []>("op_3528_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3528_cast = slice_by_index(begin = var_3528_begin_0, end = var_3528_end_0, end_mask = var_3528_end_mask_0, x = var_3500_cast)[name = tensor<string, []>("op_3528_cast")];
            tensor<int32, [4]> var_3529_begin_0 = const()[name = tensor<string, []>("op_3529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3529_end_0 = const()[name = tensor<string, []>("op_3529_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3529_end_mask_0 = const()[name = tensor<string, []>("op_3529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3529_cast = slice_by_index(begin = var_3529_begin_0, end = var_3529_end_0, end_mask = var_3529_end_mask_0, x = var_3504_cast)[name = tensor<string, []>("op_3529_cast")];
            tensor<int32, [4]> var_3530_begin_0 = const()[name = tensor<string, []>("op_3530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3530_end_0 = const()[name = tensor<string, []>("op_3530_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3530_end_mask_0 = const()[name = tensor<string, []>("op_3530_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3530_cast = slice_by_index(begin = var_3530_begin_0, end = var_3530_end_0, end_mask = var_3530_end_mask_0, x = var_3504_cast)[name = tensor<string, []>("op_3530_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, 64, 1, 512])];
            tensor<bool, [4]> var_3531_end_mask_0 = const()[name = tensor<string, []>("op_3531_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3531_cast = slice_by_index(begin = var_3531_begin_0, end = var_3531_end_0, end_mask = var_3531_end_mask_0, x = var_3508_cast)[name = tensor<string, []>("op_3531_cast")];
            tensor<int32, [4]> var_3532_begin_0 = const()[name = tensor<string, []>("op_3532_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3532_end_0 = const()[name = tensor<string, []>("op_3532_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3532_end_mask_0 = const()[name = tensor<string, []>("op_3532_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3532_cast = slice_by_index(begin = var_3532_begin_0, end = var_3532_end_0, end_mask = var_3532_end_mask_0, x = var_3508_cast)[name = tensor<string, []>("op_3532_cast")];
            tensor<int32, [4]> var_3533_begin_0 = const()[name = tensor<string, []>("op_3533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3533_end_0 = const()[name = tensor<string, []>("op_3533_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_3533_end_mask_0 = const()[name = tensor<string, []>("op_3533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3533_cast = slice_by_index(begin = var_3533_begin_0, end = var_3533_end_0, end_mask = var_3533_end_mask_0, x = var_3512_cast)[name = tensor<string, []>("op_3533_cast")];
            tensor<int32, [4]> var_3534_begin_0 = const()[name = tensor<string, []>("op_3534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_3534_end_0 = const()[name = tensor<string, []>("op_3534_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_3534_end_mask_0 = const()[name = tensor<string, []>("op_3534_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_3534_cast = slice_by_index(begin = var_3534_begin_0, end = var_3534_end_0, end_mask = var_3534_end_mask_0, x = var_3512_cast)[name = tensor<string, []>("op_3534_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_3539_begin_0 = const()[name = tensor<string, []>("op_3539_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3539_end_0 = const()[name = tensor<string, []>("op_3539_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_3539_end_mask_0 = const()[name = tensor<string, []>("op_3539_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, 64]> var_3539_cast = slice_by_index(begin = var_3539_begin_0, end = var_3539_end_0, end_mask = var_3539_end_mask_0, x = transpose_24)[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, 0, 0, 64])];
            tensor<int32, [4]> var_3543_end_0 = const()[name = tensor<string, []>("op_3543_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_3543_end_mask_0 = const()[name = tensor<string, []>("op_3543_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3543_cast = slice_by_index(begin = var_3543_begin_0, end = var_3543_end_0, end_mask = var_3543_end_mask_0, x = transpose_24)[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, 0, 0, 128])];
            tensor<int32, [4]> var_3547_end_0 = const()[name = tensor<string, []>("op_3547_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_3547_end_mask_0 = const()[name = tensor<string, []>("op_3547_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3547_cast = slice_by_index(begin = var_3547_begin_0, end = var_3547_end_0, end_mask = var_3547_end_mask_0, x = transpose_24)[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, 0, 0, 192])];
            tensor<int32, [4]> var_3551_end_0 = const()[name = tensor<string, []>("op_3551_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_3551_end_mask_0 = const()[name = tensor<string, []>("op_3551_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3551_cast = slice_by_index(begin = var_3551_begin_0, end = var_3551_end_0, end_mask = var_3551_end_mask_0, x = transpose_24)[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, 0, 0, 256])];
            tensor<int32, [4]> var_3555_end_0 = const()[name = tensor<string, []>("op_3555_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_3555_end_mask_0 = const()[name = tensor<string, []>("op_3555_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3555_cast = slice_by_index(begin = var_3555_begin_0, end = var_3555_end_0, end_mask = var_3555_end_mask_0, x = transpose_24)[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, 0, 0, 320])];
            tensor<int32, [4]> var_3559_end_0 = const()[name = tensor<string, []>("op_3559_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_3559_end_mask_0 = const()[name = tensor<string, []>("op_3559_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_3559_cast = slice_by_index(begin = var_3559_begin_0, end = var_3559_end_0, end_mask = var_3559_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_3559_cast")];
            tensor<int32, [4]> var_3563_begin_0 = const()[name = tensor<string, []>("op_3563_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_3563_end_0 = const()[name = tensor<string, []>("op_3563_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            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, 77, 1, 64]> var_3563_cast = slice_by_index(begin = var_3563_begin_0, end = var_3563_end_0, end_mask = var_3563_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_3563_cast")];
            tensor<int32, [4]> var_3567_begin_0 = const()[name = tensor<string, []>("op_3567_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_3567_end_0 = const()[name = tensor<string, []>("op_3567_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            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, 77, 1, 64]> var_3567_cast = slice_by_index(begin = var_3567_begin_0, end = var_3567_end_0, end_mask = var_3567_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_3567_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, 77, 1, 576])];
            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, 77, 1, 64]> var_3571_cast = slice_by_index(begin = var_3571_begin_0, end = var_3571_end_0, end_mask = var_3571_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_3571_cast")];
            tensor<int32, [4]> var_3575_begin_0 = const()[name = tensor<string, []>("op_3575_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_3575_end_0 = const()[name = tensor<string, []>("op_3575_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            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, 77, 1, 64]> var_3575_cast = slice_by_index(begin = var_3575_begin_0, end = var_3575_end_0, end_mask = var_3575_end_mask_0, x = transpose_24)[name = tensor<string, []>("op_3575_cast")];
            tensor<int32, [4]> var_3577_begin_0 = const()[name = tensor<string, []>("op_3577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_3577_end_0 = const()[name = tensor<string, []>("op_3577_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_3577_end_mask_0 = const()[name = tensor<string, []>("op_3577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3577_cast = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3577_cast")];
            tensor<int32, [4]> var_3581_begin_0 = const()[name = tensor<string, []>("op_3581_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_3581_end_0 = const()[name = tensor<string, []>("op_3581_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_3581_end_mask_0 = const()[name = tensor<string, []>("op_3581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3581_cast = slice_by_index(begin = var_3581_begin_0, end = var_3581_end_0, end_mask = var_3581_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3581_cast")];
            tensor<int32, [4]> var_3585_begin_0 = const()[name = tensor<string, []>("op_3585_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_3585_end_0 = const()[name = tensor<string, []>("op_3585_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_3585_end_mask_0 = const()[name = tensor<string, []>("op_3585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3585_cast = slice_by_index(begin = var_3585_begin_0, end = var_3585_end_0, end_mask = var_3585_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3585_cast")];
            tensor<int32, [4]> var_3589_begin_0 = const()[name = tensor<string, []>("op_3589_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_3589_end_0 = const()[name = tensor<string, []>("op_3589_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_3589_end_mask_0 = const()[name = tensor<string, []>("op_3589_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3589_cast = slice_by_index(begin = var_3589_begin_0, end = var_3589_end_0, end_mask = var_3589_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3589_cast")];
            tensor<int32, [4]> var_3593_begin_0 = const()[name = tensor<string, []>("op_3593_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_3593_end_0 = const()[name = tensor<string, []>("op_3593_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_3593_end_mask_0 = const()[name = tensor<string, []>("op_3593_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3593_cast = slice_by_index(begin = var_3593_begin_0, end = var_3593_end_0, end_mask = var_3593_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3593_cast")];
            tensor<int32, [4]> var_3597_begin_0 = const()[name = tensor<string, []>("op_3597_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_3597_end_0 = const()[name = tensor<string, []>("op_3597_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_3597_end_mask_0 = const()[name = tensor<string, []>("op_3597_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3597_cast = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3597_cast")];
            tensor<int32, [4]> var_3601_begin_0 = const()[name = tensor<string, []>("op_3601_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_3601_end_0 = const()[name = tensor<string, []>("op_3601_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_3601_end_mask_0 = const()[name = tensor<string, []>("op_3601_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3601_cast = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3601_cast")];
            tensor<int32, [4]> var_3605_begin_0 = const()[name = tensor<string, []>("op_3605_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_3605_end_0 = const()[name = tensor<string, []>("op_3605_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_3605_end_mask_0 = const()[name = tensor<string, []>("op_3605_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3605_cast = slice_by_index(begin = var_3605_begin_0, end = var_3605_end_0, end_mask = var_3605_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3605_cast")];
            tensor<int32, [4]> var_3609_begin_0 = const()[name = tensor<string, []>("op_3609_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_3609_end_0 = const()[name = tensor<string, []>("op_3609_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_3609_end_mask_0 = const()[name = tensor<string, []>("op_3609_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3609_cast = slice_by_index(begin = var_3609_begin_0, end = var_3609_end_0, end_mask = var_3609_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3609_cast")];
            tensor<int32, [4]> var_3613_begin_0 = const()[name = tensor<string, []>("op_3613_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_3613_end_0 = const()[name = tensor<string, []>("op_3613_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_3613_end_mask_0 = const()[name = tensor<string, []>("op_3613_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_3613_cast = slice_by_index(begin = var_3613_begin_0, end = var_3613_end_0, end_mask = var_3613_end_mask_0, x = v_15_cast)[name = tensor<string, []>("op_3613_cast")];
            tensor<string, []> var_3617_equation_0 = const()[name = tensor<string, []>("op_3617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3617_cast = einsum(equation = var_3617_equation_0, values = (var_3539_cast, var_3515_cast))[name = tensor<string, []>("op_3617_cast")];
            tensor<fp16, []> var_3618_to_fp16 = const()[name = tensor<string, []>("op_3618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_441_cast = mul(x = var_3617_cast, y = var_3618_to_fp16)[name = tensor<string, []>("aw_chunk_441_cast")];
            tensor<string, []> var_3621_equation_0 = const()[name = tensor<string, []>("op_3621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3621_cast = einsum(equation = var_3621_equation_0, values = (var_3539_cast, var_3516_cast))[name = tensor<string, []>("op_3621_cast")];
            tensor<fp16, []> var_3622_to_fp16 = const()[name = tensor<string, []>("op_3622_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_443_cast = mul(x = var_3621_cast, y = var_3622_to_fp16)[name = tensor<string, []>("aw_chunk_443_cast")];
            tensor<string, []> var_3625_equation_0 = const()[name = tensor<string, []>("op_3625_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3625_cast = einsum(equation = var_3625_equation_0, values = (var_3543_cast, var_3517_cast))[name = tensor<string, []>("op_3625_cast")];
            tensor<fp16, []> var_3626_to_fp16 = const()[name = tensor<string, []>("op_3626_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_445_cast = mul(x = var_3625_cast, y = var_3626_to_fp16)[name = tensor<string, []>("aw_chunk_445_cast")];
            tensor<string, []> var_3629_equation_0 = const()[name = tensor<string, []>("op_3629_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3629_cast = einsum(equation = var_3629_equation_0, values = (var_3543_cast, var_3518_cast))[name = tensor<string, []>("op_3629_cast")];
            tensor<fp16, []> var_3630_to_fp16 = const()[name = tensor<string, []>("op_3630_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_447_cast = mul(x = var_3629_cast, y = var_3630_to_fp16)[name = tensor<string, []>("aw_chunk_447_cast")];
            tensor<string, []> var_3633_equation_0 = const()[name = tensor<string, []>("op_3633_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3633_cast = einsum(equation = var_3633_equation_0, values = (var_3547_cast, var_3519_cast))[name = tensor<string, []>("op_3633_cast")];
            tensor<fp16, []> var_3634_to_fp16 = const()[name = tensor<string, []>("op_3634_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_449_cast = mul(x = var_3633_cast, y = var_3634_to_fp16)[name = tensor<string, []>("aw_chunk_449_cast")];
            tensor<string, []> var_3637_equation_0 = const()[name = tensor<string, []>("op_3637_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3637_cast = einsum(equation = var_3637_equation_0, values = (var_3547_cast, var_3520_cast))[name = tensor<string, []>("op_3637_cast")];
            tensor<fp16, []> var_3638_to_fp16 = const()[name = tensor<string, []>("op_3638_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_451_cast = mul(x = var_3637_cast, y = var_3638_to_fp16)[name = tensor<string, []>("aw_chunk_451_cast")];
            tensor<string, []> var_3641_equation_0 = const()[name = tensor<string, []>("op_3641_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3641_cast = einsum(equation = var_3641_equation_0, values = (var_3551_cast, var_3521_cast))[name = tensor<string, []>("op_3641_cast")];
            tensor<fp16, []> var_3642_to_fp16 = const()[name = tensor<string, []>("op_3642_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_453_cast = mul(x = var_3641_cast, y = var_3642_to_fp16)[name = tensor<string, []>("aw_chunk_453_cast")];
            tensor<string, []> var_3645_equation_0 = const()[name = tensor<string, []>("op_3645_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3645_cast = einsum(equation = var_3645_equation_0, values = (var_3551_cast, var_3522_cast))[name = tensor<string, []>("op_3645_cast")];
            tensor<fp16, []> var_3646_to_fp16 = const()[name = tensor<string, []>("op_3646_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_455_cast = mul(x = var_3645_cast, y = var_3646_to_fp16)[name = tensor<string, []>("aw_chunk_455_cast")];
            tensor<string, []> var_3649_equation_0 = const()[name = tensor<string, []>("op_3649_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3649_cast = einsum(equation = var_3649_equation_0, values = (var_3555_cast, var_3523_cast))[name = tensor<string, []>("op_3649_cast")];
            tensor<fp16, []> var_3650_to_fp16 = const()[name = tensor<string, []>("op_3650_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_457_cast = mul(x = var_3649_cast, y = var_3650_to_fp16)[name = tensor<string, []>("aw_chunk_457_cast")];
            tensor<string, []> var_3653_equation_0 = const()[name = tensor<string, []>("op_3653_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3653_cast = einsum(equation = var_3653_equation_0, values = (var_3555_cast, var_3524_cast))[name = tensor<string, []>("op_3653_cast")];
            tensor<fp16, []> var_3654_to_fp16 = const()[name = tensor<string, []>("op_3654_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_459_cast = mul(x = var_3653_cast, y = var_3654_to_fp16)[name = tensor<string, []>("aw_chunk_459_cast")];
            tensor<string, []> var_3657_equation_0 = const()[name = tensor<string, []>("op_3657_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3657_cast = einsum(equation = var_3657_equation_0, values = (var_3559_cast, var_3525_cast))[name = tensor<string, []>("op_3657_cast")];
            tensor<fp16, []> var_3658_to_fp16 = const()[name = tensor<string, []>("op_3658_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_461_cast = mul(x = var_3657_cast, y = var_3658_to_fp16)[name = tensor<string, []>("aw_chunk_461_cast")];
            tensor<string, []> var_3661_equation_0 = const()[name = tensor<string, []>("op_3661_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3661_cast = einsum(equation = var_3661_equation_0, values = (var_3559_cast, var_3526_cast))[name = tensor<string, []>("op_3661_cast")];
            tensor<fp16, []> var_3662_to_fp16 = const()[name = tensor<string, []>("op_3662_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_463_cast = mul(x = var_3661_cast, y = var_3662_to_fp16)[name = tensor<string, []>("aw_chunk_463_cast")];
            tensor<string, []> var_3665_equation_0 = const()[name = tensor<string, []>("op_3665_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3665_cast = einsum(equation = var_3665_equation_0, values = (var_3563_cast, var_3527_cast))[name = tensor<string, []>("op_3665_cast")];
            tensor<fp16, []> var_3666_to_fp16 = const()[name = tensor<string, []>("op_3666_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_465_cast = mul(x = var_3665_cast, y = var_3666_to_fp16)[name = tensor<string, []>("aw_chunk_465_cast")];
            tensor<string, []> var_3669_equation_0 = const()[name = tensor<string, []>("op_3669_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3669_cast = einsum(equation = var_3669_equation_0, values = (var_3563_cast, var_3528_cast))[name = tensor<string, []>("op_3669_cast")];
            tensor<fp16, []> var_3670_to_fp16 = const()[name = tensor<string, []>("op_3670_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_467_cast = mul(x = var_3669_cast, y = var_3670_to_fp16)[name = tensor<string, []>("aw_chunk_467_cast")];
            tensor<string, []> var_3673_equation_0 = const()[name = tensor<string, []>("op_3673_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3673_cast = einsum(equation = var_3673_equation_0, values = (var_3567_cast, var_3529_cast))[name = tensor<string, []>("op_3673_cast")];
            tensor<fp16, []> var_3674_to_fp16 = const()[name = tensor<string, []>("op_3674_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_469_cast = mul(x = var_3673_cast, y = var_3674_to_fp16)[name = tensor<string, []>("aw_chunk_469_cast")];
            tensor<string, []> var_3677_equation_0 = const()[name = tensor<string, []>("op_3677_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3677_cast = einsum(equation = var_3677_equation_0, values = (var_3567_cast, var_3530_cast))[name = tensor<string, []>("op_3677_cast")];
            tensor<fp16, []> var_3678_to_fp16 = const()[name = tensor<string, []>("op_3678_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_471_cast = mul(x = var_3677_cast, y = var_3678_to_fp16)[name = tensor<string, []>("aw_chunk_471_cast")];
            tensor<string, []> var_3681_equation_0 = const()[name = tensor<string, []>("op_3681_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3681_cast = einsum(equation = var_3681_equation_0, values = (var_3571_cast, var_3531_cast))[name = tensor<string, []>("op_3681_cast")];
            tensor<fp16, []> var_3682_to_fp16 = const()[name = tensor<string, []>("op_3682_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_473_cast = mul(x = var_3681_cast, y = var_3682_to_fp16)[name = tensor<string, []>("aw_chunk_473_cast")];
            tensor<string, []> var_3685_equation_0 = const()[name = tensor<string, []>("op_3685_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3685_cast = einsum(equation = var_3685_equation_0, values = (var_3571_cast, var_3532_cast))[name = tensor<string, []>("op_3685_cast")];
            tensor<fp16, []> var_3686_to_fp16 = const()[name = tensor<string, []>("op_3686_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_475_cast = mul(x = var_3685_cast, y = var_3686_to_fp16)[name = tensor<string, []>("aw_chunk_475_cast")];
            tensor<string, []> var_3689_equation_0 = const()[name = tensor<string, []>("op_3689_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3689_cast = einsum(equation = var_3689_equation_0, values = (var_3575_cast, var_3533_cast))[name = tensor<string, []>("op_3689_cast")];
            tensor<fp16, []> var_3690_to_fp16 = const()[name = tensor<string, []>("op_3690_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_477_cast = mul(x = var_3689_cast, y = var_3690_to_fp16)[name = tensor<string, []>("aw_chunk_477_cast")];
            tensor<string, []> var_3693_equation_0 = const()[name = tensor<string, []>("op_3693_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_3693_cast = einsum(equation = var_3693_equation_0, values = (var_3575_cast, var_3534_cast))[name = tensor<string, []>("op_3693_cast")];
            tensor<fp16, []> var_3694_to_fp16 = const()[name = tensor<string, []>("op_3694_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_479_cast = mul(x = var_3693_cast, y = var_3694_to_fp16)[name = tensor<string, []>("aw_chunk_479_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3696_cast = softmax(axis = var_2173, x = aw_chunk_441_cast)[name = tensor<string, []>("op_3696_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3697_cast = softmax(axis = var_2173, x = aw_chunk_443_cast)[name = tensor<string, []>("op_3697_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3698_cast = softmax(axis = var_2173, x = aw_chunk_445_cast)[name = tensor<string, []>("op_3698_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3699_cast = softmax(axis = var_2173, x = aw_chunk_447_cast)[name = tensor<string, []>("op_3699_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3700_cast = softmax(axis = var_2173, x = aw_chunk_449_cast)[name = tensor<string, []>("op_3700_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3701_cast = softmax(axis = var_2173, x = aw_chunk_451_cast)[name = tensor<string, []>("op_3701_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3702_cast = softmax(axis = var_2173, x = aw_chunk_453_cast)[name = tensor<string, []>("op_3702_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3703_cast = softmax(axis = var_2173, x = aw_chunk_455_cast)[name = tensor<string, []>("op_3703_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3704_cast = softmax(axis = var_2173, x = aw_chunk_457_cast)[name = tensor<string, []>("op_3704_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3705_cast = softmax(axis = var_2173, x = aw_chunk_459_cast)[name = tensor<string, []>("op_3705_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3706_cast = softmax(axis = var_2173, x = aw_chunk_461_cast)[name = tensor<string, []>("op_3706_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3707_cast = softmax(axis = var_2173, x = aw_chunk_463_cast)[name = tensor<string, []>("op_3707_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3708_cast = softmax(axis = var_2173, x = aw_chunk_465_cast)[name = tensor<string, []>("op_3708_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3709_cast = softmax(axis = var_2173, x = aw_chunk_467_cast)[name = tensor<string, []>("op_3709_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3710_cast = softmax(axis = var_2173, x = aw_chunk_469_cast)[name = tensor<string, []>("op_3710_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3711_cast = softmax(axis = var_2173, x = aw_chunk_471_cast)[name = tensor<string, []>("op_3711_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3712_cast = softmax(axis = var_2173, x = aw_chunk_473_cast)[name = tensor<string, []>("op_3712_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3713_cast = softmax(axis = var_2173, x = aw_chunk_475_cast)[name = tensor<string, []>("op_3713_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3714_cast = softmax(axis = var_2173, x = aw_chunk_477_cast)[name = tensor<string, []>("op_3714_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_3715_cast = softmax(axis = var_2173, x = aw_chunk_479_cast)[name = tensor<string, []>("op_3715_cast")];
            tensor<string, []> var_3717_equation_0 = const()[name = tensor<string, []>("op_3717_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3717_cast = einsum(equation = var_3717_equation_0, values = (var_3577_cast, var_3696_cast))[name = tensor<string, []>("op_3717_cast")];
            tensor<string, []> var_3719_equation_0 = const()[name = tensor<string, []>("op_3719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3719_cast = einsum(equation = var_3719_equation_0, values = (var_3577_cast, var_3697_cast))[name = tensor<string, []>("op_3719_cast")];
            tensor<string, []> var_3721_equation_0 = const()[name = tensor<string, []>("op_3721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3721_cast = einsum(equation = var_3721_equation_0, values = (var_3581_cast, var_3698_cast))[name = tensor<string, []>("op_3721_cast")];
            tensor<string, []> var_3723_equation_0 = const()[name = tensor<string, []>("op_3723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3723_cast = einsum(equation = var_3723_equation_0, values = (var_3581_cast, var_3699_cast))[name = tensor<string, []>("op_3723_cast")];
            tensor<string, []> var_3725_equation_0 = const()[name = tensor<string, []>("op_3725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3725_cast = einsum(equation = var_3725_equation_0, values = (var_3585_cast, var_3700_cast))[name = tensor<string, []>("op_3725_cast")];
            tensor<string, []> var_3727_equation_0 = const()[name = tensor<string, []>("op_3727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3727_cast = einsum(equation = var_3727_equation_0, values = (var_3585_cast, var_3701_cast))[name = tensor<string, []>("op_3727_cast")];
            tensor<string, []> var_3729_equation_0 = const()[name = tensor<string, []>("op_3729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3729_cast = einsum(equation = var_3729_equation_0, values = (var_3589_cast, var_3702_cast))[name = tensor<string, []>("op_3729_cast")];
            tensor<string, []> var_3731_equation_0 = const()[name = tensor<string, []>("op_3731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3731_cast = einsum(equation = var_3731_equation_0, values = (var_3589_cast, var_3703_cast))[name = tensor<string, []>("op_3731_cast")];
            tensor<string, []> var_3733_equation_0 = const()[name = tensor<string, []>("op_3733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3733_cast = einsum(equation = var_3733_equation_0, values = (var_3593_cast, var_3704_cast))[name = tensor<string, []>("op_3733_cast")];
            tensor<string, []> var_3735_equation_0 = const()[name = tensor<string, []>("op_3735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3735_cast = einsum(equation = var_3735_equation_0, values = (var_3593_cast, var_3705_cast))[name = tensor<string, []>("op_3735_cast")];
            tensor<string, []> var_3737_equation_0 = const()[name = tensor<string, []>("op_3737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3737_cast = einsum(equation = var_3737_equation_0, values = (var_3597_cast, var_3706_cast))[name = tensor<string, []>("op_3737_cast")];
            tensor<string, []> var_3739_equation_0 = const()[name = tensor<string, []>("op_3739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3739_cast = einsum(equation = var_3739_equation_0, values = (var_3597_cast, var_3707_cast))[name = tensor<string, []>("op_3739_cast")];
            tensor<string, []> var_3741_equation_0 = const()[name = tensor<string, []>("op_3741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3741_cast = einsum(equation = var_3741_equation_0, values = (var_3601_cast, var_3708_cast))[name = tensor<string, []>("op_3741_cast")];
            tensor<string, []> var_3743_equation_0 = const()[name = tensor<string, []>("op_3743_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3743_cast = einsum(equation = var_3743_equation_0, values = (var_3601_cast, var_3709_cast))[name = tensor<string, []>("op_3743_cast")];
            tensor<string, []> var_3745_equation_0 = const()[name = tensor<string, []>("op_3745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3745_cast = einsum(equation = var_3745_equation_0, values = (var_3605_cast, var_3710_cast))[name = tensor<string, []>("op_3745_cast")];
            tensor<string, []> var_3747_equation_0 = const()[name = tensor<string, []>("op_3747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3747_cast = einsum(equation = var_3747_equation_0, values = (var_3605_cast, var_3711_cast))[name = tensor<string, []>("op_3747_cast")];
            tensor<string, []> var_3749_equation_0 = const()[name = tensor<string, []>("op_3749_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3749_cast = einsum(equation = var_3749_equation_0, values = (var_3609_cast, var_3712_cast))[name = tensor<string, []>("op_3749_cast")];
            tensor<string, []> var_3751_equation_0 = const()[name = tensor<string, []>("op_3751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3751_cast = einsum(equation = var_3751_equation_0, values = (var_3609_cast, var_3713_cast))[name = tensor<string, []>("op_3751_cast")];
            tensor<string, []> var_3753_equation_0 = const()[name = tensor<string, []>("op_3753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3753_cast = einsum(equation = var_3753_equation_0, values = (var_3613_cast, var_3714_cast))[name = tensor<string, []>("op_3753_cast")];
            tensor<string, []> var_3755_equation_0 = const()[name = tensor<string, []>("op_3755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_3755_cast = einsum(equation = var_3755_equation_0, values = (var_3613_cast, var_3715_cast))[name = tensor<string, []>("op_3755_cast")];
            tensor<bool, []> var_3757_interleave_0 = const()[name = tensor<string, []>("op_3757_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3757_cast = concat(axis = var_2150, interleave = var_3757_interleave_0, values = (var_3717_cast, var_3719_cast))[name = tensor<string, []>("op_3757_cast")];
            tensor<bool, []> var_3759_interleave_0 = const()[name = tensor<string, []>("op_3759_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3759_cast = concat(axis = var_2150, interleave = var_3759_interleave_0, values = (var_3721_cast, var_3723_cast))[name = tensor<string, []>("op_3759_cast")];
            tensor<bool, []> var_3761_interleave_0 = const()[name = tensor<string, []>("op_3761_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3761_cast = concat(axis = var_2150, interleave = var_3761_interleave_0, values = (var_3725_cast, var_3727_cast))[name = tensor<string, []>("op_3761_cast")];
            tensor<bool, []> var_3763_interleave_0 = const()[name = tensor<string, []>("op_3763_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3763_cast = concat(axis = var_2150, interleave = var_3763_interleave_0, values = (var_3729_cast, var_3731_cast))[name = tensor<string, []>("op_3763_cast")];
            tensor<bool, []> var_3765_interleave_0 = const()[name = tensor<string, []>("op_3765_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3765_cast = concat(axis = var_2150, interleave = var_3765_interleave_0, values = (var_3733_cast, var_3735_cast))[name = tensor<string, []>("op_3765_cast")];
            tensor<bool, []> var_3767_interleave_0 = const()[name = tensor<string, []>("op_3767_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3767_cast = concat(axis = var_2150, interleave = var_3767_interleave_0, values = (var_3737_cast, var_3739_cast))[name = tensor<string, []>("op_3767_cast")];
            tensor<bool, []> var_3769_interleave_0 = const()[name = tensor<string, []>("op_3769_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3769_cast = concat(axis = var_2150, interleave = var_3769_interleave_0, values = (var_3741_cast, var_3743_cast))[name = tensor<string, []>("op_3769_cast")];
            tensor<bool, []> var_3771_interleave_0 = const()[name = tensor<string, []>("op_3771_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3771_cast = concat(axis = var_2150, interleave = var_3771_interleave_0, values = (var_3745_cast, var_3747_cast))[name = tensor<string, []>("op_3771_cast")];
            tensor<bool, []> var_3773_interleave_0 = const()[name = tensor<string, []>("op_3773_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3773_cast = concat(axis = var_2150, interleave = var_3773_interleave_0, values = (var_3749_cast, var_3751_cast))[name = tensor<string, []>("op_3773_cast")];
            tensor<bool, []> var_3775_interleave_0 = const()[name = tensor<string, []>("op_3775_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_3775_cast = concat(axis = var_2150, interleave = var_3775_interleave_0, values = (var_3753_cast, var_3755_cast))[name = tensor<string, []>("op_3775_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_2173, interleave = input_107_interleave_0, values = (var_3757_cast, var_3759_cast, var_3761_cast, var_3763_cast, var_3765_cast, var_3767_cast, var_3769_cast, var_3771_cast, var_3773_cast, var_3775_cast))[name = tensor<string, []>("input_107_cast")];
            tensor<int32, [2]> var_3781 = const()[name = tensor<string, []>("op_3781"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3783 = const()[name = tensor<string, []>("op_3783"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3785_pad_type_0 = const()[name = tensor<string, []>("op_3785_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3785_pad_0 = const()[name = tensor<string, []>("op_3785_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10320640))), lut = tensor<fp16, [4]>([-0x1.714p-6, -0x1.508p-8, 0x1.508p-8, 0x1.718p-6]), 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, []>(10423104)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3785_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_3783, groups = var_2173, pad = var_3785_pad_0, pad_type = var_3785_pad_type_0, strides = var_3781, 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_3785_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_23_cast = add(x = var_3785_cast, y = inputs_21_cast)[name = tensor<string, []>("inputs_23_cast")];
            tensor<int32, [1]> var_3789 = const()[name = tensor<string, []>("op_3789"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_23_cast = reduce_mean(axes = var_3789, keep_dims = var_2168, 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_3793 = const()[name = tensor<string, []>("op_3793"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_3794_cast = reduce_mean(axes = var_3793, keep_dims = var_2168, x = zero_mean_sq_23_cast)[name = tensor<string, []>("op_3794_cast")];
            tensor<fp16, []> var_3795_to_fp16 = const()[name = tensor<string, []>("op_3795_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_3796_cast = add(x = var_3794_cast, y = var_3795_to_fp16)[name = tensor<string, []>("op_3796_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_3796_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_3800_to_fp16 = const()[name = tensor<string, []>("op_3800_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10424448)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3801_cast = add(x = out_23_cast, y = var_3800_to_fp16)[name = tensor<string, []>("op_3801_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_3803_to_fp16 = const()[name = tensor<string, []>("op_3803_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10425792)))];
            tensor<fp16, [2, 640, 1, 1024]> input_109_cast = mul(x = var_3801_cast, y = var_3803_to_fp16)[name = tensor<string, []>("input_109_cast")];
            tensor<int32, [2]> var_3811 = const()[name = tensor<string, []>("op_3811"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3813 = const()[name = tensor<string, []>("op_3813"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3815_pad_type_0 = const()[name = tensor<string, []>("op_3815_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3815_pad_0 = const()[name = tensor<string, []>("op_3815_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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10427136))), lut = tensor<fp16, [4]>([-0x1.89p-4, -0x1.cfp-6, 0x1.d24p-6, 0x1.8ap-4]), 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11246400))), lut = tensor<fp16, [4]>([-0x1.328p-6, 0x1.548p-5, -0x1.a28p-4, -0x1.17cp-2]), 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_3815_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_3813, groups = var_2173, pad = var_3815_pad_0, pad_type = var_3815_pad_type_0, strides = var_3811, 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_3815_cast")];
            tensor<int32, [2]> var_3816_split_sizes_0 = const()[name = tensor<string, []>("op_3816_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_3816_axis_0 = const()[name = tensor<string, []>("op_3816_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_3816_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_3816_cast_1 = split(axis = var_3816_axis_0, split_sizes = var_3816_split_sizes_0, x = var_3815_cast)[name = tensor<string, []>("op_3816_cast")];
            tensor<string, []> var_3818_mode_0 = const()[name = tensor<string, []>("op_3818_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_3818_cast = gelu(mode = var_3818_mode_0, x = var_3816_cast_1)[name = tensor<string, []>("op_3818_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_111_cast = mul(x = var_3816_cast_0, y = var_3818_cast)[name = tensor<string, []>("input_111_cast")];
            tensor<int32, [2]> var_3822 = const()[name = tensor<string, []>("op_3822"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3824 = const()[name = tensor<string, []>("op_3824"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_3826_pad_type_0 = const()[name = tensor<string, []>("op_3826_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_3826_pad_0 = const()[name = tensor<string, []>("op_3826_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11247744))), lut = tensor<fp16, [4]>([-0x1.98p-4, -0x1.e6cp-6, 0x1.e7cp-6, 0x1.99p-4]), 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, []>(11657408)))];
            tensor<fp16, [2, 640, 1, 1024]> var_3826_cast = conv(bias = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_3824, groups = var_2173, pad = var_3826_pad_0, pad_type = var_3826_pad_type_0, strides = var_3822, weight = down_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_111_cast)[name = tensor<string, []>("op_3826_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_69_cast = add(x = var_3826_cast, y = inputs_23_cast)[name = tensor<string, []>("hidden_states_69_cast")];
            tensor<int32, [4]> var_3828 = const()[name = tensor<string, []>("op_3828"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_113_cast = reshape(shape = var_3828, x = hidden_states_69_cast)[name = tensor<string, []>("input_113_cast")];
            tensor<int32, [2]> var_3832 = const()[name = tensor<string, []>("op_3832"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3834 = const()[name = tensor<string, []>("op_3834"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11658752))), lut = tensor<fp16, [4]>([-0x1.7b8p-4, -0x1.c9cp-6, 0x1.c2cp-6, 0x1.798p-4]), 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, []>(11761216)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_71_cast = conv(bias = down_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_3834, groups = var_2173, pad = hidden_states_71_pad_0, pad_type = hidden_states_71_pad_type_0, strides = var_3832, 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_3841 = const()[name = tensor<string, []>("op_3841"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_3843 = const()[name = tensor<string, []>("op_3843"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11762560))), lut = tensor<fp16, [4]>([-0x1.48p-5, -0x1.5dp-7, 0x1.7c4p-7, 0x1.574p-5]), 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, []>(12684224)))];
            tensor<fp16, [2, 640, 16, 16]> input_117_cast = conv(bias = down_blocks_1_downsamplers_0_conv_bias_to_fp16, dilations = var_3843, groups = var_2173, pad = input_117_pad_0, pad_type = input_117_pad_type_0, strides = var_3841, weight = down_blocks_1_downsamplers_0_conv_weight_to_fp16_palettized, x = input_115_cast)[name = tensor<string, []>("input_117_cast")];
            tensor<bool, []> var_3878 = const()[name = tensor<string, []>("op_3878"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_3883 = const()[name = tensor<string, []>("op_3883"), 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, []>(12685568)))];
            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, []>(12686912)))];
            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_3906 = const()[name = tensor<string, []>("op_3906"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3908 = const()[name = tensor<string, []>("op_3908"), 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, [1843200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12688256))), lut = tensor<fp16, [4]>([-0x1.62p-4, -0x1.8bcp-6, 0x1.6a8p-6, 0x1.59cp-4]), 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, []>(14531520)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_73_cast = conv(bias = down_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_3908, groups = var_3883, pad = hidden_states_73_pad_0, pad_type = hidden_states_73_pad_type_0, strides = var_3906, 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_3914 = const()[name = tensor<string, []>("op_3914"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3916 = const()[name = tensor<string, []>("op_3916"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14534144))), lut = tensor<fp16, [4]>([-0x1.23p+2, -0x1.3p-8, 0x1.d6cp-9, 0x1.7a4p-5]), 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, []>(14943808)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_9_cast = conv(bias = down_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_3916, groups = var_3883, pad = temb_9_pad_0, pad_type = temb_9_pad_type_0, strides = var_3914, 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, []>(14946432)))];
            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, []>(14949056)))];
            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, []>(14951680)))];
            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, []>(14954304)))];
            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_3926 = const()[name = tensor<string, []>("op_3926"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3928 = const()[name = tensor<string, []>("op_3928"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14956928))), lut = tensor<fp16, [4]>([-0x1.1fp-4, -0x1.46p-6, 0x1.454p-6, 0x1.1f4p-4]), 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, []>(18643392)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_75_cast = conv(bias = down_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_3928, groups = var_3883, pad = hidden_states_75_pad_0, pad_type = hidden_states_75_pad_type_0, strides = var_3926, 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_3933 = const()[name = tensor<string, []>("op_3933"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3935 = const()[name = tensor<string, []>("op_3935"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18646016))), lut = tensor<fp16, [4]>([-0x1.ee4p-6, -0x1.164p-7, 0x1.2fcp-7, 0x1.fc4p-6]), 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, []>(18850880)))];
            tensor<fp16, [2, 1280, 16, 16]> x_3_cast = conv(bias = down_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_3935, groups = var_3883, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = var_3933, 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, []>(18853504)))];
            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, []>(18856128)))];
            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_3955 = const()[name = tensor<string, []>("op_3955"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3957 = const()[name = tensor<string, []>("op_3957"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18858752))), lut = tensor<fp16, [4]>([-0x1.8cp-4, -0x1.dbp-6, 0x1.dc8p-6, 0x1.8d4p-4]), 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, []>(19268416)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_79_cast = conv(bias = down_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_3957, groups = var_3883, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = var_3955, 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_3962 = const()[name = tensor<string, []>("op_3962"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_25_cast = reshape(shape = var_3962, x = hidden_states_79_cast)[name = tensor<string, []>("inputs_25_cast")];
            tensor<int32, [1]> var_3972 = const()[name = tensor<string, []>("op_3972"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_25_cast = reduce_mean(axes = var_3972, keep_dims = var_3878, 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_3976 = const()[name = tensor<string, []>("op_3976"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_3977_cast = reduce_mean(axes = var_3976, keep_dims = var_3878, x = zero_mean_sq_25_cast)[name = tensor<string, []>("op_3977_cast")];
            tensor<fp16, []> var_3978_to_fp16 = const()[name = tensor<string, []>("op_3978_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_3979_cast = add(x = var_3977_cast, y = var_3978_to_fp16)[name = tensor<string, []>("op_3979_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_3979_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_3983_to_fp16 = const()[name = tensor<string, []>("op_3983_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19271040)))];
            tensor<fp16, [2, 1280, 1, 256]> var_3984_cast = add(x = out_25_cast, y = var_3983_to_fp16)[name = tensor<string, []>("op_3984_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_3986_to_fp16 = const()[name = tensor<string, []>("op_3986_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19273664)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_81_cast = mul(x = var_3984_cast, y = var_3986_to_fp16)[name = tensor<string, []>("hidden_states_81_cast")];
            tensor<int32, [2]> var_3993 = const()[name = tensor<string, []>("op_3993"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_3995 = const()[name = tensor<string, []>("op_3995"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19276288))), lut = tensor<fp16, [4]>([-0x1.91p-4, -0x1.dbp-6, 0x1.d9p-6, 0x1.9p-4]), 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_3995, groups = var_3883, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = var_3993, 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_3999 = const()[name = tensor<string, []>("op_3999"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4001 = const()[name = tensor<string, []>("op_4001"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19685952))), lut = tensor<fp16, [4]>([-0x1.90cp-4, -0x1.d98p-6, 0x1.e08p-6, 0x1.928p-4]), 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_4001, groups = var_3883, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = var_3999, 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_4005 = const()[name = tensor<string, []>("op_4005"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4007 = const()[name = tensor<string, []>("op_4007"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20095616))), lut = tensor<fp16, [4]>([-0x1.308p-4, -0x1.6b4p-6, 0x1.694p-6, 0x1.2fp-4]), 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_4007, groups = var_3883, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = var_4005, 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_4011_begin_0 = const()[name = tensor<string, []>("op_4011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4011_end_0 = const()[name = tensor<string, []>("op_4011_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_4011_end_mask_0 = const()[name = tensor<string, []>("op_4011_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4011_cast = slice_by_index(begin = var_4011_begin_0, end = var_4011_end_0, end_mask = var_4011_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4011_cast")];
            tensor<int32, [4]> var_4015_begin_0 = const()[name = tensor<string, []>("op_4015_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4015_end_0 = const()[name = tensor<string, []>("op_4015_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_4015_end_mask_0 = const()[name = tensor<string, []>("op_4015_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4015_cast = slice_by_index(begin = var_4015_begin_0, end = var_4015_end_0, end_mask = var_4015_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4015_cast")];
            tensor<int32, [4]> var_4019_begin_0 = const()[name = tensor<string, []>("op_4019_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4019_end_0 = const()[name = tensor<string, []>("op_4019_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_4019_end_mask_0 = const()[name = tensor<string, []>("op_4019_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4019_cast = slice_by_index(begin = var_4019_begin_0, end = var_4019_end_0, end_mask = var_4019_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4019_cast")];
            tensor<int32, [4]> var_4023_begin_0 = const()[name = tensor<string, []>("op_4023_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4023_end_0 = const()[name = tensor<string, []>("op_4023_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            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, 64, 1, 256]> var_4023_cast = slice_by_index(begin = var_4023_begin_0, end = var_4023_end_0, end_mask = var_4023_end_mask_0, x = q_17_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, 256, 0, 0])];
            tensor<int32, [4]> var_4027_end_0 = const()[name = tensor<string, []>("op_4027_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            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, 64, 1, 256]> var_4027_cast = slice_by_index(begin = var_4027_begin_0, end = var_4027_end_0, end_mask = var_4027_end_mask_0, x = q_17_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, 320, 0, 0])];
            tensor<int32, [4]> var_4031_end_0 = const()[name = tensor<string, []>("op_4031_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            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, 64, 1, 256]> var_4031_cast = slice_by_index(begin = var_4031_begin_0, end = var_4031_end_0, end_mask = var_4031_end_mask_0, x = q_17_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, 384, 0, 0])];
            tensor<int32, [4]> var_4035_end_0 = const()[name = tensor<string, []>("op_4035_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            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, 64, 1, 256]> var_4035_cast = slice_by_index(begin = var_4035_begin_0, end = var_4035_end_0, end_mask = var_4035_end_mask_0, x = q_17_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, 448, 0, 0])];
            tensor<int32, [4]> var_4039_end_0 = const()[name = tensor<string, []>("op_4039_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            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, 64, 1, 256]> var_4039_cast = slice_by_index(begin = var_4039_begin_0, end = var_4039_end_0, end_mask = var_4039_end_mask_0, x = q_17_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, 512, 0, 0])];
            tensor<int32, [4]> var_4043_end_0 = const()[name = tensor<string, []>("op_4043_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            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, 64, 1, 256]> var_4043_cast = slice_by_index(begin = var_4043_begin_0, end = var_4043_end_0, end_mask = var_4043_end_mask_0, x = q_17_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, 576, 0, 0])];
            tensor<int32, [4]> var_4047_end_0 = const()[name = tensor<string, []>("op_4047_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            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, 64, 1, 256]> var_4047_cast = slice_by_index(begin = var_4047_begin_0, end = var_4047_end_0, end_mask = var_4047_end_mask_0, x = q_17_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, 640, 0, 0])];
            tensor<int32, [4]> var_4051_end_0 = const()[name = tensor<string, []>("op_4051_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            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, 64, 1, 256]> var_4051_cast = slice_by_index(begin = var_4051_begin_0, end = var_4051_end_0, end_mask = var_4051_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4051_cast")];
            tensor<int32, [4]> var_4055_begin_0 = const()[name = tensor<string, []>("op_4055_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4055_end_0 = const()[name = tensor<string, []>("op_4055_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_4055_end_mask_0 = const()[name = tensor<string, []>("op_4055_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4055_cast = slice_by_index(begin = var_4055_begin_0, end = var_4055_end_0, end_mask = var_4055_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4055_cast")];
            tensor<int32, [4]> var_4059_begin_0 = const()[name = tensor<string, []>("op_4059_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4059_end_0 = const()[name = tensor<string, []>("op_4059_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_4059_end_mask_0 = const()[name = tensor<string, []>("op_4059_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4059_cast = slice_by_index(begin = var_4059_begin_0, end = var_4059_end_0, end_mask = var_4059_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4059_cast")];
            tensor<int32, [4]> var_4063_begin_0 = const()[name = tensor<string, []>("op_4063_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4063_end_0 = const()[name = tensor<string, []>("op_4063_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_4063_end_mask_0 = const()[name = tensor<string, []>("op_4063_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4063_cast = slice_by_index(begin = var_4063_begin_0, end = var_4063_end_0, end_mask = var_4063_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4063_cast")];
            tensor<int32, [4]> var_4067_begin_0 = const()[name = tensor<string, []>("op_4067_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4067_end_0 = const()[name = tensor<string, []>("op_4067_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_4067_end_mask_0 = const()[name = tensor<string, []>("op_4067_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4067_cast = slice_by_index(begin = var_4067_begin_0, end = var_4067_end_0, end_mask = var_4067_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4067_cast")];
            tensor<int32, [4]> var_4071_begin_0 = const()[name = tensor<string, []>("op_4071_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4071_end_0 = const()[name = tensor<string, []>("op_4071_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_4071_end_mask_0 = const()[name = tensor<string, []>("op_4071_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4071_cast = slice_by_index(begin = var_4071_begin_0, end = var_4071_end_0, end_mask = var_4071_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4071_cast")];
            tensor<int32, [4]> var_4075_begin_0 = const()[name = tensor<string, []>("op_4075_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4075_end_0 = const()[name = tensor<string, []>("op_4075_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_4075_end_mask_0 = const()[name = tensor<string, []>("op_4075_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4075_cast = slice_by_index(begin = var_4075_begin_0, end = var_4075_end_0, end_mask = var_4075_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4075_cast")];
            tensor<int32, [4]> var_4079_begin_0 = const()[name = tensor<string, []>("op_4079_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4079_end_0 = const()[name = tensor<string, []>("op_4079_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_4079_end_mask_0 = const()[name = tensor<string, []>("op_4079_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4079_cast = slice_by_index(begin = var_4079_begin_0, end = var_4079_end_0, end_mask = var_4079_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4079_cast")];
            tensor<int32, [4]> var_4083_begin_0 = const()[name = tensor<string, []>("op_4083_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4083_end_0 = const()[name = tensor<string, []>("op_4083_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_4083_end_mask_0 = const()[name = tensor<string, []>("op_4083_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4083_cast = slice_by_index(begin = var_4083_begin_0, end = var_4083_end_0, end_mask = var_4083_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4083_cast")];
            tensor<int32, [4]> var_4087_begin_0 = const()[name = tensor<string, []>("op_4087_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4087_end_0 = const()[name = tensor<string, []>("op_4087_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_4087_end_mask_0 = const()[name = tensor<string, []>("op_4087_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4087_cast = slice_by_index(begin = var_4087_begin_0, end = var_4087_end_0, end_mask = var_4087_end_mask_0, x = q_17_cast)[name = tensor<string, []>("op_4087_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_4094_begin_0 = const()[name = tensor<string, []>("op_4094_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4094_end_0 = const()[name = tensor<string, []>("op_4094_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_4094_end_mask_0 = const()[name = tensor<string, []>("op_4094_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, 64]> var_4094_cast = slice_by_index(begin = var_4094_begin_0, end = var_4094_end_0, end_mask = var_4094_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4094_cast")];
            tensor<int32, [4]> var_4098_begin_0 = const()[name = tensor<string, []>("op_4098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_4098_end_0 = const()[name = tensor<string, []>("op_4098_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_4098_end_mask_0 = const()[name = tensor<string, []>("op_4098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4098_cast = slice_by_index(begin = var_4098_begin_0, end = var_4098_end_0, end_mask = var_4098_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4098_cast")];
            tensor<int32, [4]> var_4102_begin_0 = const()[name = tensor<string, []>("op_4102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_4102_end_0 = const()[name = tensor<string, []>("op_4102_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_4102_end_mask_0 = const()[name = tensor<string, []>("op_4102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4102_cast = slice_by_index(begin = var_4102_begin_0, end = var_4102_end_0, end_mask = var_4102_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4102_cast")];
            tensor<int32, [4]> var_4106_begin_0 = const()[name = tensor<string, []>("op_4106_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_4106_end_0 = const()[name = tensor<string, []>("op_4106_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_4106_end_mask_0 = const()[name = tensor<string, []>("op_4106_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4106_cast = slice_by_index(begin = var_4106_begin_0, end = var_4106_end_0, end_mask = var_4106_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4106_cast")];
            tensor<int32, [4]> var_4110_begin_0 = const()[name = tensor<string, []>("op_4110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_4110_end_0 = const()[name = tensor<string, []>("op_4110_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_4110_end_mask_0 = const()[name = tensor<string, []>("op_4110_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4110_cast = slice_by_index(begin = var_4110_begin_0, end = var_4110_end_0, end_mask = var_4110_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4110_cast")];
            tensor<int32, [4]> var_4114_begin_0 = const()[name = tensor<string, []>("op_4114_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4114_end_0 = const()[name = tensor<string, []>("op_4114_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_4114_end_mask_0 = const()[name = tensor<string, []>("op_4114_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4114_cast = slice_by_index(begin = var_4114_begin_0, end = var_4114_end_0, end_mask = var_4114_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4114_cast")];
            tensor<int32, [4]> var_4118_begin_0 = const()[name = tensor<string, []>("op_4118_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_4118_end_0 = const()[name = tensor<string, []>("op_4118_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_4118_end_mask_0 = const()[name = tensor<string, []>("op_4118_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4118_cast = slice_by_index(begin = var_4118_begin_0, end = var_4118_end_0, end_mask = var_4118_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4118_cast")];
            tensor<int32, [4]> var_4122_begin_0 = const()[name = tensor<string, []>("op_4122_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_4122_end_0 = const()[name = tensor<string, []>("op_4122_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_4122_end_mask_0 = const()[name = tensor<string, []>("op_4122_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4122_cast = slice_by_index(begin = var_4122_begin_0, end = var_4122_end_0, end_mask = var_4122_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4122_cast")];
            tensor<int32, [4]> var_4126_begin_0 = const()[name = tensor<string, []>("op_4126_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4126_end_0 = const()[name = tensor<string, []>("op_4126_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_4126_end_mask_0 = const()[name = tensor<string, []>("op_4126_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4126_cast = slice_by_index(begin = var_4126_begin_0, end = var_4126_end_0, end_mask = var_4126_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4126_cast")];
            tensor<int32, [4]> var_4130_begin_0 = const()[name = tensor<string, []>("op_4130_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_4130_end_0 = const()[name = tensor<string, []>("op_4130_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_4130_end_mask_0 = const()[name = tensor<string, []>("op_4130_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4130_cast = slice_by_index(begin = var_4130_begin_0, end = var_4130_end_0, end_mask = var_4130_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4130_cast")];
            tensor<int32, [4]> var_4134_begin_0 = const()[name = tensor<string, []>("op_4134_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4134_end_0 = const()[name = tensor<string, []>("op_4134_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_4134_end_mask_0 = const()[name = tensor<string, []>("op_4134_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4134_cast = slice_by_index(begin = var_4134_begin_0, end = var_4134_end_0, end_mask = var_4134_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4134_cast")];
            tensor<int32, [4]> var_4138_begin_0 = const()[name = tensor<string, []>("op_4138_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_4138_end_0 = const()[name = tensor<string, []>("op_4138_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_4138_end_mask_0 = const()[name = tensor<string, []>("op_4138_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4138_cast = slice_by_index(begin = var_4138_begin_0, end = var_4138_end_0, end_mask = var_4138_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4138_cast")];
            tensor<int32, [4]> var_4142_begin_0 = const()[name = tensor<string, []>("op_4142_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_4142_end_0 = const()[name = tensor<string, []>("op_4142_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_4142_end_mask_0 = const()[name = tensor<string, []>("op_4142_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4142_cast = slice_by_index(begin = var_4142_begin_0, end = var_4142_end_0, end_mask = var_4142_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4142_cast")];
            tensor<int32, [4]> var_4146_begin_0 = const()[name = tensor<string, []>("op_4146_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_4146_end_0 = const()[name = tensor<string, []>("op_4146_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_4146_end_mask_0 = const()[name = tensor<string, []>("op_4146_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4146_cast = slice_by_index(begin = var_4146_begin_0, end = var_4146_end_0, end_mask = var_4146_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4146_cast")];
            tensor<int32, [4]> var_4150_begin_0 = const()[name = tensor<string, []>("op_4150_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_4150_end_0 = const()[name = tensor<string, []>("op_4150_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_4150_end_mask_0 = const()[name = tensor<string, []>("op_4150_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4150_cast = slice_by_index(begin = var_4150_begin_0, end = var_4150_end_0, end_mask = var_4150_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4150_cast")];
            tensor<int32, [4]> var_4154_begin_0 = const()[name = tensor<string, []>("op_4154_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4154_end_0 = const()[name = tensor<string, []>("op_4154_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_4154_end_mask_0 = const()[name = tensor<string, []>("op_4154_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4154_cast = slice_by_index(begin = var_4154_begin_0, end = var_4154_end_0, end_mask = var_4154_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4154_cast")];
            tensor<int32, [4]> var_4158_begin_0 = const()[name = tensor<string, []>("op_4158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_4158_end_0 = const()[name = tensor<string, []>("op_4158_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_4158_end_mask_0 = const()[name = tensor<string, []>("op_4158_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4158_cast = slice_by_index(begin = var_4158_begin_0, end = var_4158_end_0, end_mask = var_4158_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4158_cast")];
            tensor<int32, [4]> var_4162_begin_0 = const()[name = tensor<string, []>("op_4162_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_4162_end_0 = const()[name = tensor<string, []>("op_4162_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_4162_end_mask_0 = const()[name = tensor<string, []>("op_4162_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4162_cast = slice_by_index(begin = var_4162_begin_0, end = var_4162_end_0, end_mask = var_4162_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4162_cast")];
            tensor<int32, [4]> var_4166_begin_0 = const()[name = tensor<string, []>("op_4166_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_4166_end_0 = const()[name = tensor<string, []>("op_4166_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_4166_end_mask_0 = const()[name = tensor<string, []>("op_4166_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4166_cast = slice_by_index(begin = var_4166_begin_0, end = var_4166_end_0, end_mask = var_4166_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4166_cast")];
            tensor<int32, [4]> var_4170_begin_0 = const()[name = tensor<string, []>("op_4170_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_4170_end_0 = const()[name = tensor<string, []>("op_4170_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_4170_end_mask_0 = const()[name = tensor<string, []>("op_4170_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_4170_cast = slice_by_index(begin = var_4170_begin_0, end = var_4170_end_0, end_mask = var_4170_end_mask_0, x = transpose_23)[name = tensor<string, []>("op_4170_cast")];
            tensor<int32, [4]> var_4172_begin_0 = const()[name = tensor<string, []>("op_4172_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4172_end_0 = const()[name = tensor<string, []>("op_4172_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_4172_end_mask_0 = const()[name = tensor<string, []>("op_4172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4172_cast = slice_by_index(begin = var_4172_begin_0, end = var_4172_end_0, end_mask = var_4172_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4172_cast")];
            tensor<int32, [4]> var_4176_begin_0 = const()[name = tensor<string, []>("op_4176_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4176_end_0 = const()[name = tensor<string, []>("op_4176_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_4176_end_mask_0 = const()[name = tensor<string, []>("op_4176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4176_cast = slice_by_index(begin = var_4176_begin_0, end = var_4176_end_0, end_mask = var_4176_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4176_cast")];
            tensor<int32, [4]> var_4180_begin_0 = const()[name = tensor<string, []>("op_4180_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4180_end_0 = const()[name = tensor<string, []>("op_4180_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_4180_end_mask_0 = const()[name = tensor<string, []>("op_4180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4180_cast = slice_by_index(begin = var_4180_begin_0, end = var_4180_end_0, end_mask = var_4180_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4180_cast")];
            tensor<int32, [4]> var_4184_begin_0 = const()[name = tensor<string, []>("op_4184_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4184_end_0 = const()[name = tensor<string, []>("op_4184_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_4184_end_mask_0 = const()[name = tensor<string, []>("op_4184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4184_cast = slice_by_index(begin = var_4184_begin_0, end = var_4184_end_0, end_mask = var_4184_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4184_cast")];
            tensor<int32, [4]> var_4188_begin_0 = const()[name = tensor<string, []>("op_4188_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4188_end_0 = const()[name = tensor<string, []>("op_4188_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_4188_end_mask_0 = const()[name = tensor<string, []>("op_4188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4188_cast = slice_by_index(begin = var_4188_begin_0, end = var_4188_end_0, end_mask = var_4188_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4188_cast")];
            tensor<int32, [4]> var_4192_begin_0 = const()[name = tensor<string, []>("op_4192_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4192_end_0 = const()[name = tensor<string, []>("op_4192_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_4192_end_mask_0 = const()[name = tensor<string, []>("op_4192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4192_cast = slice_by_index(begin = var_4192_begin_0, end = var_4192_end_0, end_mask = var_4192_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4192_cast")];
            tensor<int32, [4]> var_4196_begin_0 = const()[name = tensor<string, []>("op_4196_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4196_end_0 = const()[name = tensor<string, []>("op_4196_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_4196_end_mask_0 = const()[name = tensor<string, []>("op_4196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4196_cast = slice_by_index(begin = var_4196_begin_0, end = var_4196_end_0, end_mask = var_4196_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4196_cast")];
            tensor<int32, [4]> var_4200_begin_0 = const()[name = tensor<string, []>("op_4200_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4200_end_0 = const()[name = tensor<string, []>("op_4200_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_4200_end_mask_0 = const()[name = tensor<string, []>("op_4200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4200_cast = slice_by_index(begin = var_4200_begin_0, end = var_4200_end_0, end_mask = var_4200_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4200_cast")];
            tensor<int32, [4]> var_4204_begin_0 = const()[name = tensor<string, []>("op_4204_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4204_end_0 = const()[name = tensor<string, []>("op_4204_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_4204_end_mask_0 = const()[name = tensor<string, []>("op_4204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4204_cast = slice_by_index(begin = var_4204_begin_0, end = var_4204_end_0, end_mask = var_4204_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4204_cast")];
            tensor<int32, [4]> var_4208_begin_0 = const()[name = tensor<string, []>("op_4208_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4208_end_0 = const()[name = tensor<string, []>("op_4208_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_4208_end_mask_0 = const()[name = tensor<string, []>("op_4208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4208_cast = slice_by_index(begin = var_4208_begin_0, end = var_4208_end_0, end_mask = var_4208_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4208_cast")];
            tensor<int32, [4]> var_4212_begin_0 = const()[name = tensor<string, []>("op_4212_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4212_end_0 = const()[name = tensor<string, []>("op_4212_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_4212_end_mask_0 = const()[name = tensor<string, []>("op_4212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4212_cast = slice_by_index(begin = var_4212_begin_0, end = var_4212_end_0, end_mask = var_4212_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4212_cast")];
            tensor<int32, [4]> var_4216_begin_0 = const()[name = tensor<string, []>("op_4216_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4216_end_0 = const()[name = tensor<string, []>("op_4216_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_4216_end_mask_0 = const()[name = tensor<string, []>("op_4216_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4216_cast = slice_by_index(begin = var_4216_begin_0, end = var_4216_end_0, end_mask = var_4216_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4216_cast")];
            tensor<int32, [4]> var_4220_begin_0 = const()[name = tensor<string, []>("op_4220_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4220_end_0 = const()[name = tensor<string, []>("op_4220_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_4220_end_mask_0 = const()[name = tensor<string, []>("op_4220_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4220_cast = slice_by_index(begin = var_4220_begin_0, end = var_4220_end_0, end_mask = var_4220_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4220_cast")];
            tensor<int32, [4]> var_4224_begin_0 = const()[name = tensor<string, []>("op_4224_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4224_end_0 = const()[name = tensor<string, []>("op_4224_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_4224_end_mask_0 = const()[name = tensor<string, []>("op_4224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4224_cast = slice_by_index(begin = var_4224_begin_0, end = var_4224_end_0, end_mask = var_4224_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4224_cast")];
            tensor<int32, [4]> var_4228_begin_0 = const()[name = tensor<string, []>("op_4228_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4228_end_0 = const()[name = tensor<string, []>("op_4228_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_4228_end_mask_0 = const()[name = tensor<string, []>("op_4228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4228_cast = slice_by_index(begin = var_4228_begin_0, end = var_4228_end_0, end_mask = var_4228_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4228_cast")];
            tensor<int32, [4]> var_4232_begin_0 = const()[name = tensor<string, []>("op_4232_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4232_end_0 = const()[name = tensor<string, []>("op_4232_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_4232_end_mask_0 = const()[name = tensor<string, []>("op_4232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4232_cast = slice_by_index(begin = var_4232_begin_0, end = var_4232_end_0, end_mask = var_4232_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4232_cast")];
            tensor<int32, [4]> var_4236_begin_0 = const()[name = tensor<string, []>("op_4236_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4236_end_0 = const()[name = tensor<string, []>("op_4236_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_4236_end_mask_0 = const()[name = tensor<string, []>("op_4236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4236_cast = slice_by_index(begin = var_4236_begin_0, end = var_4236_end_0, end_mask = var_4236_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4236_cast")];
            tensor<int32, [4]> var_4240_begin_0 = const()[name = tensor<string, []>("op_4240_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4240_end_0 = const()[name = tensor<string, []>("op_4240_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_4240_end_mask_0 = const()[name = tensor<string, []>("op_4240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4240_cast = slice_by_index(begin = var_4240_begin_0, end = var_4240_end_0, end_mask = var_4240_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4240_cast")];
            tensor<int32, [4]> var_4244_begin_0 = const()[name = tensor<string, []>("op_4244_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4244_end_0 = const()[name = tensor<string, []>("op_4244_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_4244_end_mask_0 = const()[name = tensor<string, []>("op_4244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4244_cast = slice_by_index(begin = var_4244_begin_0, end = var_4244_end_0, end_mask = var_4244_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4244_cast")];
            tensor<int32, [4]> var_4248_begin_0 = const()[name = tensor<string, []>("op_4248_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4248_end_0 = const()[name = tensor<string, []>("op_4248_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_4248_end_mask_0 = const()[name = tensor<string, []>("op_4248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4248_cast = slice_by_index(begin = var_4248_begin_0, end = var_4248_end_0, end_mask = var_4248_end_mask_0, x = v_17_cast)[name = tensor<string, []>("op_4248_cast")];
            tensor<string, []> var_4252_equation_0 = const()[name = tensor<string, []>("op_4252_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4252_cast = einsum(equation = var_4252_equation_0, values = (var_4094_cast, var_4011_cast))[name = tensor<string, []>("op_4252_cast")];
            tensor<fp16, []> var_4253_to_fp16 = const()[name = tensor<string, []>("op_4253_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_1_cast = mul(x = var_4252_cast, y = var_4253_to_fp16)[name = tensor<string, []>("aw_1_cast")];
            tensor<string, []> var_4256_equation_0 = const()[name = tensor<string, []>("op_4256_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4256_cast = einsum(equation = var_4256_equation_0, values = (var_4098_cast, var_4015_cast))[name = tensor<string, []>("op_4256_cast")];
            tensor<fp16, []> var_4257_to_fp16 = const()[name = tensor<string, []>("op_4257_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_3_cast = mul(x = var_4256_cast, y = var_4257_to_fp16)[name = tensor<string, []>("aw_3_cast")];
            tensor<string, []> var_4260_equation_0 = const()[name = tensor<string, []>("op_4260_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4260_cast = einsum(equation = var_4260_equation_0, values = (var_4102_cast, var_4019_cast))[name = tensor<string, []>("op_4260_cast")];
            tensor<fp16, []> var_4261_to_fp16 = const()[name = tensor<string, []>("op_4261_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_5_cast = mul(x = var_4260_cast, y = var_4261_to_fp16)[name = tensor<string, []>("aw_5_cast")];
            tensor<string, []> var_4264_equation_0 = const()[name = tensor<string, []>("op_4264_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4264_cast = einsum(equation = var_4264_equation_0, values = (var_4106_cast, var_4023_cast))[name = tensor<string, []>("op_4264_cast")];
            tensor<fp16, []> var_4265_to_fp16 = const()[name = tensor<string, []>("op_4265_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_7_cast = mul(x = var_4264_cast, y = var_4265_to_fp16)[name = tensor<string, []>("aw_7_cast")];
            tensor<string, []> var_4268_equation_0 = const()[name = tensor<string, []>("op_4268_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4268_cast = einsum(equation = var_4268_equation_0, values = (var_4110_cast, var_4027_cast))[name = tensor<string, []>("op_4268_cast")];
            tensor<fp16, []> var_4269_to_fp16 = const()[name = tensor<string, []>("op_4269_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_9_cast = mul(x = var_4268_cast, y = var_4269_to_fp16)[name = tensor<string, []>("aw_9_cast")];
            tensor<string, []> var_4272_equation_0 = const()[name = tensor<string, []>("op_4272_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4272_cast = einsum(equation = var_4272_equation_0, values = (var_4114_cast, var_4031_cast))[name = tensor<string, []>("op_4272_cast")];
            tensor<fp16, []> var_4273_to_fp16 = const()[name = tensor<string, []>("op_4273_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_11_cast = mul(x = var_4272_cast, y = var_4273_to_fp16)[name = tensor<string, []>("aw_11_cast")];
            tensor<string, []> var_4276_equation_0 = const()[name = tensor<string, []>("op_4276_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4276_cast = einsum(equation = var_4276_equation_0, values = (var_4118_cast, var_4035_cast))[name = tensor<string, []>("op_4276_cast")];
            tensor<fp16, []> var_4277_to_fp16 = const()[name = tensor<string, []>("op_4277_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_13_cast = mul(x = var_4276_cast, y = var_4277_to_fp16)[name = tensor<string, []>("aw_13_cast")];
            tensor<string, []> var_4280_equation_0 = const()[name = tensor<string, []>("op_4280_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4280_cast = einsum(equation = var_4280_equation_0, values = (var_4122_cast, var_4039_cast))[name = tensor<string, []>("op_4280_cast")];
            tensor<fp16, []> var_4281_to_fp16 = const()[name = tensor<string, []>("op_4281_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_15_cast = mul(x = var_4280_cast, y = var_4281_to_fp16)[name = tensor<string, []>("aw_15_cast")];
            tensor<string, []> var_4284_equation_0 = const()[name = tensor<string, []>("op_4284_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4284_cast = einsum(equation = var_4284_equation_0, values = (var_4126_cast, var_4043_cast))[name = tensor<string, []>("op_4284_cast")];
            tensor<fp16, []> var_4285_to_fp16 = const()[name = tensor<string, []>("op_4285_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_17_cast = mul(x = var_4284_cast, y = var_4285_to_fp16)[name = tensor<string, []>("aw_17_cast")];
            tensor<string, []> var_4288_equation_0 = const()[name = tensor<string, []>("op_4288_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4288_cast = einsum(equation = var_4288_equation_0, values = (var_4130_cast, var_4047_cast))[name = tensor<string, []>("op_4288_cast")];
            tensor<fp16, []> var_4289_to_fp16 = const()[name = tensor<string, []>("op_4289_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_19_cast = mul(x = var_4288_cast, y = var_4289_to_fp16)[name = tensor<string, []>("aw_19_cast")];
            tensor<string, []> var_4292_equation_0 = const()[name = tensor<string, []>("op_4292_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4292_cast = einsum(equation = var_4292_equation_0, values = (var_4134_cast, var_4051_cast))[name = tensor<string, []>("op_4292_cast")];
            tensor<fp16, []> var_4293_to_fp16 = const()[name = tensor<string, []>("op_4293_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_21_cast = mul(x = var_4292_cast, y = var_4293_to_fp16)[name = tensor<string, []>("aw_21_cast")];
            tensor<string, []> var_4296_equation_0 = const()[name = tensor<string, []>("op_4296_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4296_cast = einsum(equation = var_4296_equation_0, values = (var_4138_cast, var_4055_cast))[name = tensor<string, []>("op_4296_cast")];
            tensor<fp16, []> var_4297_to_fp16 = const()[name = tensor<string, []>("op_4297_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_23_cast = mul(x = var_4296_cast, y = var_4297_to_fp16)[name = tensor<string, []>("aw_23_cast")];
            tensor<string, []> var_4300_equation_0 = const()[name = tensor<string, []>("op_4300_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4300_cast = einsum(equation = var_4300_equation_0, values = (var_4142_cast, var_4059_cast))[name = tensor<string, []>("op_4300_cast")];
            tensor<fp16, []> var_4301_to_fp16 = const()[name = tensor<string, []>("op_4301_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_25_cast = mul(x = var_4300_cast, y = var_4301_to_fp16)[name = tensor<string, []>("aw_25_cast")];
            tensor<string, []> var_4304_equation_0 = const()[name = tensor<string, []>("op_4304_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4304_cast = einsum(equation = var_4304_equation_0, values = (var_4146_cast, var_4063_cast))[name = tensor<string, []>("op_4304_cast")];
            tensor<fp16, []> var_4305_to_fp16 = const()[name = tensor<string, []>("op_4305_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_27_cast = mul(x = var_4304_cast, y = var_4305_to_fp16)[name = tensor<string, []>("aw_27_cast")];
            tensor<string, []> var_4308_equation_0 = const()[name = tensor<string, []>("op_4308_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4308_cast = einsum(equation = var_4308_equation_0, values = (var_4150_cast, var_4067_cast))[name = tensor<string, []>("op_4308_cast")];
            tensor<fp16, []> var_4309_to_fp16 = const()[name = tensor<string, []>("op_4309_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_29_cast = mul(x = var_4308_cast, y = var_4309_to_fp16)[name = tensor<string, []>("aw_29_cast")];
            tensor<string, []> var_4312_equation_0 = const()[name = tensor<string, []>("op_4312_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4312_cast = einsum(equation = var_4312_equation_0, values = (var_4154_cast, var_4071_cast))[name = tensor<string, []>("op_4312_cast")];
            tensor<fp16, []> var_4313_to_fp16 = const()[name = tensor<string, []>("op_4313_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_31_cast = mul(x = var_4312_cast, y = var_4313_to_fp16)[name = tensor<string, []>("aw_31_cast")];
            tensor<string, []> var_4316_equation_0 = const()[name = tensor<string, []>("op_4316_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4316_cast = einsum(equation = var_4316_equation_0, values = (var_4158_cast, var_4075_cast))[name = tensor<string, []>("op_4316_cast")];
            tensor<fp16, []> var_4317_to_fp16 = const()[name = tensor<string, []>("op_4317_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_33_cast = mul(x = var_4316_cast, y = var_4317_to_fp16)[name = tensor<string, []>("aw_33_cast")];
            tensor<string, []> var_4320_equation_0 = const()[name = tensor<string, []>("op_4320_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4320_cast = einsum(equation = var_4320_equation_0, values = (var_4162_cast, var_4079_cast))[name = tensor<string, []>("op_4320_cast")];
            tensor<fp16, []> var_4321_to_fp16 = const()[name = tensor<string, []>("op_4321_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_35_cast = mul(x = var_4320_cast, y = var_4321_to_fp16)[name = tensor<string, []>("aw_35_cast")];
            tensor<string, []> var_4324_equation_0 = const()[name = tensor<string, []>("op_4324_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4324_cast = einsum(equation = var_4324_equation_0, values = (var_4166_cast, var_4083_cast))[name = tensor<string, []>("op_4324_cast")];
            tensor<fp16, []> var_4325_to_fp16 = const()[name = tensor<string, []>("op_4325_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_37_cast = mul(x = var_4324_cast, y = var_4325_to_fp16)[name = tensor<string, []>("aw_37_cast")];
            tensor<string, []> var_4328_equation_0 = const()[name = tensor<string, []>("op_4328_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_4328_cast = einsum(equation = var_4328_equation_0, values = (var_4170_cast, var_4087_cast))[name = tensor<string, []>("op_4328_cast")];
            tensor<fp16, []> var_4329_to_fp16 = const()[name = tensor<string, []>("op_4329_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_39_cast = mul(x = var_4328_cast, y = var_4329_to_fp16)[name = tensor<string, []>("aw_39_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4331_cast = softmax(axis = var_3883, x = aw_1_cast)[name = tensor<string, []>("op_4331_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4332_cast = softmax(axis = var_3883, x = aw_3_cast)[name = tensor<string, []>("op_4332_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4333_cast = softmax(axis = var_3883, x = aw_5_cast)[name = tensor<string, []>("op_4333_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4334_cast = softmax(axis = var_3883, x = aw_7_cast)[name = tensor<string, []>("op_4334_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4335_cast = softmax(axis = var_3883, x = aw_9_cast)[name = tensor<string, []>("op_4335_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4336_cast = softmax(axis = var_3883, x = aw_11_cast)[name = tensor<string, []>("op_4336_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4337_cast = softmax(axis = var_3883, x = aw_13_cast)[name = tensor<string, []>("op_4337_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4338_cast = softmax(axis = var_3883, x = aw_15_cast)[name = tensor<string, []>("op_4338_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4339_cast = softmax(axis = var_3883, x = aw_17_cast)[name = tensor<string, []>("op_4339_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4340_cast = softmax(axis = var_3883, x = aw_19_cast)[name = tensor<string, []>("op_4340_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4341_cast = softmax(axis = var_3883, x = aw_21_cast)[name = tensor<string, []>("op_4341_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4342_cast = softmax(axis = var_3883, x = aw_23_cast)[name = tensor<string, []>("op_4342_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4343_cast = softmax(axis = var_3883, x = aw_25_cast)[name = tensor<string, []>("op_4343_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4344_cast = softmax(axis = var_3883, x = aw_27_cast)[name = tensor<string, []>("op_4344_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4345_cast = softmax(axis = var_3883, x = aw_29_cast)[name = tensor<string, []>("op_4345_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4346_cast = softmax(axis = var_3883, x = aw_31_cast)[name = tensor<string, []>("op_4346_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4347_cast = softmax(axis = var_3883, x = aw_33_cast)[name = tensor<string, []>("op_4347_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4348_cast = softmax(axis = var_3883, x = aw_35_cast)[name = tensor<string, []>("op_4348_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4349_cast = softmax(axis = var_3883, x = aw_37_cast)[name = tensor<string, []>("op_4349_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_4350_cast = softmax(axis = var_3883, x = aw_39_cast)[name = tensor<string, []>("op_4350_cast")];
            tensor<string, []> var_4352_equation_0 = const()[name = tensor<string, []>("op_4352_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4352_cast = einsum(equation = var_4352_equation_0, values = (var_4172_cast, var_4331_cast))[name = tensor<string, []>("op_4352_cast")];
            tensor<string, []> var_4354_equation_0 = const()[name = tensor<string, []>("op_4354_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4354_cast = einsum(equation = var_4354_equation_0, values = (var_4176_cast, var_4332_cast))[name = tensor<string, []>("op_4354_cast")];
            tensor<string, []> var_4356_equation_0 = const()[name = tensor<string, []>("op_4356_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4356_cast = einsum(equation = var_4356_equation_0, values = (var_4180_cast, var_4333_cast))[name = tensor<string, []>("op_4356_cast")];
            tensor<string, []> var_4358_equation_0 = const()[name = tensor<string, []>("op_4358_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4358_cast = einsum(equation = var_4358_equation_0, values = (var_4184_cast, var_4334_cast))[name = tensor<string, []>("op_4358_cast")];
            tensor<string, []> var_4360_equation_0 = const()[name = tensor<string, []>("op_4360_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4360_cast = einsum(equation = var_4360_equation_0, values = (var_4188_cast, var_4335_cast))[name = tensor<string, []>("op_4360_cast")];
            tensor<string, []> var_4362_equation_0 = const()[name = tensor<string, []>("op_4362_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4362_cast = einsum(equation = var_4362_equation_0, values = (var_4192_cast, var_4336_cast))[name = tensor<string, []>("op_4362_cast")];
            tensor<string, []> var_4364_equation_0 = const()[name = tensor<string, []>("op_4364_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4364_cast = einsum(equation = var_4364_equation_0, values = (var_4196_cast, var_4337_cast))[name = tensor<string, []>("op_4364_cast")];
            tensor<string, []> var_4366_equation_0 = const()[name = tensor<string, []>("op_4366_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4366_cast = einsum(equation = var_4366_equation_0, values = (var_4200_cast, var_4338_cast))[name = tensor<string, []>("op_4366_cast")];
            tensor<string, []> var_4368_equation_0 = const()[name = tensor<string, []>("op_4368_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4368_cast = einsum(equation = var_4368_equation_0, values = (var_4204_cast, var_4339_cast))[name = tensor<string, []>("op_4368_cast")];
            tensor<string, []> var_4370_equation_0 = const()[name = tensor<string, []>("op_4370_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4370_cast = einsum(equation = var_4370_equation_0, values = (var_4208_cast, var_4340_cast))[name = tensor<string, []>("op_4370_cast")];
            tensor<string, []> var_4372_equation_0 = const()[name = tensor<string, []>("op_4372_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4372_cast = einsum(equation = var_4372_equation_0, values = (var_4212_cast, var_4341_cast))[name = tensor<string, []>("op_4372_cast")];
            tensor<string, []> var_4374_equation_0 = const()[name = tensor<string, []>("op_4374_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4374_cast = einsum(equation = var_4374_equation_0, values = (var_4216_cast, var_4342_cast))[name = tensor<string, []>("op_4374_cast")];
            tensor<string, []> var_4376_equation_0 = const()[name = tensor<string, []>("op_4376_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4376_cast = einsum(equation = var_4376_equation_0, values = (var_4220_cast, var_4343_cast))[name = tensor<string, []>("op_4376_cast")];
            tensor<string, []> var_4378_equation_0 = const()[name = tensor<string, []>("op_4378_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4378_cast = einsum(equation = var_4378_equation_0, values = (var_4224_cast, var_4344_cast))[name = tensor<string, []>("op_4378_cast")];
            tensor<string, []> var_4380_equation_0 = const()[name = tensor<string, []>("op_4380_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4380_cast = einsum(equation = var_4380_equation_0, values = (var_4228_cast, var_4345_cast))[name = tensor<string, []>("op_4380_cast")];
            tensor<string, []> var_4382_equation_0 = const()[name = tensor<string, []>("op_4382_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4382_cast = einsum(equation = var_4382_equation_0, values = (var_4232_cast, var_4346_cast))[name = tensor<string, []>("op_4382_cast")];
            tensor<string, []> var_4384_equation_0 = const()[name = tensor<string, []>("op_4384_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4384_cast = einsum(equation = var_4384_equation_0, values = (var_4236_cast, var_4347_cast))[name = tensor<string, []>("op_4384_cast")];
            tensor<string, []> var_4386_equation_0 = const()[name = tensor<string, []>("op_4386_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4386_cast = einsum(equation = var_4386_equation_0, values = (var_4240_cast, var_4348_cast))[name = tensor<string, []>("op_4386_cast")];
            tensor<string, []> var_4388_equation_0 = const()[name = tensor<string, []>("op_4388_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4388_cast = einsum(equation = var_4388_equation_0, values = (var_4244_cast, var_4349_cast))[name = tensor<string, []>("op_4388_cast")];
            tensor<string, []> var_4390_equation_0 = const()[name = tensor<string, []>("op_4390_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4390_cast = einsum(equation = var_4390_equation_0, values = (var_4248_cast, var_4350_cast))[name = tensor<string, []>("op_4390_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_3883, interleave = input_133_interleave_0, values = (var_4352_cast, var_4354_cast, var_4356_cast, var_4358_cast, var_4360_cast, var_4362_cast, var_4364_cast, var_4366_cast, var_4368_cast, var_4370_cast, var_4372_cast, var_4374_cast, var_4376_cast, var_4378_cast, var_4380_cast, var_4382_cast, var_4384_cast, var_4386_cast, var_4388_cast, var_4390_cast))[name = tensor<string, []>("input_133_cast")];
            tensor<int32, [2]> var_4396 = const()[name = tensor<string, []>("op_4396"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4398 = const()[name = tensor<string, []>("op_4398"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4400_pad_type_0 = const()[name = tensor<string, []>("op_4400_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4400_pad_0 = const()[name = tensor<string, []>("op_4400_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20505280))), lut = tensor<fp16, [4]>([-0x1.4ecp-4, -0x1.8e8p-6, 0x1.95cp-6, 0x1.518p-4]), 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, []>(20914944)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4400_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_4398, groups = var_3883, pad = var_4400_pad_0, pad_type = var_4400_pad_type_0, strides = var_4396, 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_4400_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_27_cast = add(x = var_4400_cast, y = inputs_25_cast)[name = tensor<string, []>("inputs_27_cast")];
            tensor<int32, [1]> var_4404 = const()[name = tensor<string, []>("op_4404"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_27_cast = reduce_mean(axes = var_4404, keep_dims = var_3878, 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_4408 = const()[name = tensor<string, []>("op_4408"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_4409_cast = reduce_mean(axes = var_4408, keep_dims = var_3878, x = zero_mean_sq_27_cast)[name = tensor<string, []>("op_4409_cast")];
            tensor<fp16, []> var_4410_to_fp16 = const()[name = tensor<string, []>("op_4410_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_4411_cast = add(x = var_4409_cast, y = var_4410_to_fp16)[name = tensor<string, []>("op_4411_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_4411_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_4415_to_fp16 = const()[name = tensor<string, []>("op_4415_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20917568)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4416_cast = add(x = out_27_cast, y = var_4415_to_fp16)[name = tensor<string, []>("op_4416_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4418_to_fp16 = const()[name = tensor<string, []>("op_4418_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20920192)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_83_cast = mul(x = var_4416_cast, y = var_4418_to_fp16)[name = tensor<string, []>("hidden_states_83_cast")];
            tensor<int32, [2]> var_4425 = const()[name = tensor<string, []>("op_4425"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4427 = const()[name = tensor<string, []>("op_4427"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20922816))), lut = tensor<fp16, [4]>([-0x1.348p-4, -0x1.6c8p-6, 0x1.6f8p-6, 0x1.34cp-4]), 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_4427, groups = var_3883, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = var_4425, 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_4431 = const()[name = tensor<string, []>("op_4431"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4433 = const()[name = tensor<string, []>("op_4433"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_37_pad_type_0 = const()[name = tensor<string, []>("k_37_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_37_pad_0 = const()[name = tensor<string, []>("k_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21332480))), lut = tensor<fp16, [4]>([-0x1.43p-4, -0x1.7d8p-6, 0x1.7ap-6, 0x1.428p-4]), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_37_cast = conv(dilations = var_4433, groups = var_3883, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = var_4431, 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_4437 = const()[name = tensor<string, []>("op_4437"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4439 = const()[name = tensor<string, []>("op_4439"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_19_pad_type_0 = const()[name = tensor<string, []>("v_19_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_19_pad_0 = const()[name = tensor<string, []>("v_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21660224))), lut = tensor<fp16, [4]>([-0x1.a1p-5, -0x1.ebp-7, 0x1.f04p-7, 0x1.a3p-5]), name = tensor<string, []>("down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_19_cast = conv(dilations = var_4439, groups = var_3883, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = var_4437, 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_4443_begin_0 = const()[name = tensor<string, []>("op_4443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4443_end_0 = const()[name = tensor<string, []>("op_4443_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_4443_end_mask_0 = const()[name = tensor<string, []>("op_4443_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4443_cast = slice_by_index(begin = var_4443_begin_0, end = var_4443_end_0, end_mask = var_4443_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4443_cast")];
            tensor<int32, [4]> var_4447_begin_0 = const()[name = tensor<string, []>("op_4447_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4447_end_0 = const()[name = tensor<string, []>("op_4447_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_4447_end_mask_0 = const()[name = tensor<string, []>("op_4447_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4447_cast = slice_by_index(begin = var_4447_begin_0, end = var_4447_end_0, end_mask = var_4447_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4447_cast")];
            tensor<int32, [4]> var_4451_begin_0 = const()[name = tensor<string, []>("op_4451_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4451_end_0 = const()[name = tensor<string, []>("op_4451_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_4451_end_mask_0 = const()[name = tensor<string, []>("op_4451_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4451_cast = slice_by_index(begin = var_4451_begin_0, end = var_4451_end_0, end_mask = var_4451_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4451_cast")];
            tensor<int32, [4]> var_4455_begin_0 = const()[name = tensor<string, []>("op_4455_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4455_end_0 = const()[name = tensor<string, []>("op_4455_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_4455_end_mask_0 = const()[name = tensor<string, []>("op_4455_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4455_cast = slice_by_index(begin = var_4455_begin_0, end = var_4455_end_0, end_mask = var_4455_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4455_cast")];
            tensor<int32, [4]> var_4459_begin_0 = const()[name = tensor<string, []>("op_4459_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4459_end_0 = const()[name = tensor<string, []>("op_4459_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_4459_end_mask_0 = const()[name = tensor<string, []>("op_4459_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4459_cast = slice_by_index(begin = var_4459_begin_0, end = var_4459_end_0, end_mask = var_4459_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4459_cast")];
            tensor<int32, [4]> var_4463_begin_0 = const()[name = tensor<string, []>("op_4463_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4463_end_0 = const()[name = tensor<string, []>("op_4463_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_4463_end_mask_0 = const()[name = tensor<string, []>("op_4463_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4463_cast = slice_by_index(begin = var_4463_begin_0, end = var_4463_end_0, end_mask = var_4463_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4463_cast")];
            tensor<int32, [4]> var_4467_begin_0 = const()[name = tensor<string, []>("op_4467_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4467_end_0 = const()[name = tensor<string, []>("op_4467_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_4467_end_mask_0 = const()[name = tensor<string, []>("op_4467_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4467_cast = slice_by_index(begin = var_4467_begin_0, end = var_4467_end_0, end_mask = var_4467_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4467_cast")];
            tensor<int32, [4]> var_4471_begin_0 = const()[name = tensor<string, []>("op_4471_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4471_end_0 = const()[name = tensor<string, []>("op_4471_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_4471_end_mask_0 = const()[name = tensor<string, []>("op_4471_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4471_cast = slice_by_index(begin = var_4471_begin_0, end = var_4471_end_0, end_mask = var_4471_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4471_cast")];
            tensor<int32, [4]> var_4475_begin_0 = const()[name = tensor<string, []>("op_4475_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4475_end_0 = const()[name = tensor<string, []>("op_4475_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_4475_end_mask_0 = const()[name = tensor<string, []>("op_4475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4475_cast = slice_by_index(begin = var_4475_begin_0, end = var_4475_end_0, end_mask = var_4475_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4475_cast")];
            tensor<int32, [4]> var_4479_begin_0 = const()[name = tensor<string, []>("op_4479_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4479_end_0 = const()[name = tensor<string, []>("op_4479_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_4479_end_mask_0 = const()[name = tensor<string, []>("op_4479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4479_cast = slice_by_index(begin = var_4479_begin_0, end = var_4479_end_0, end_mask = var_4479_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4479_cast")];
            tensor<int32, [4]> var_4483_begin_0 = const()[name = tensor<string, []>("op_4483_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4483_end_0 = const()[name = tensor<string, []>("op_4483_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_4483_end_mask_0 = const()[name = tensor<string, []>("op_4483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4483_cast = slice_by_index(begin = var_4483_begin_0, end = var_4483_end_0, end_mask = var_4483_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4483_cast")];
            tensor<int32, [4]> var_4487_begin_0 = const()[name = tensor<string, []>("op_4487_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4487_end_0 = const()[name = tensor<string, []>("op_4487_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_4487_end_mask_0 = const()[name = tensor<string, []>("op_4487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4487_cast = slice_by_index(begin = var_4487_begin_0, end = var_4487_end_0, end_mask = var_4487_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4487_cast")];
            tensor<int32, [4]> var_4491_begin_0 = const()[name = tensor<string, []>("op_4491_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4491_end_0 = const()[name = tensor<string, []>("op_4491_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_4491_end_mask_0 = const()[name = tensor<string, []>("op_4491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4491_cast = slice_by_index(begin = var_4491_begin_0, end = var_4491_end_0, end_mask = var_4491_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4491_cast")];
            tensor<int32, [4]> var_4495_begin_0 = const()[name = tensor<string, []>("op_4495_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4495_end_0 = const()[name = tensor<string, []>("op_4495_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_4495_end_mask_0 = const()[name = tensor<string, []>("op_4495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4495_cast = slice_by_index(begin = var_4495_begin_0, end = var_4495_end_0, end_mask = var_4495_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4495_cast")];
            tensor<int32, [4]> var_4499_begin_0 = const()[name = tensor<string, []>("op_4499_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4499_end_0 = const()[name = tensor<string, []>("op_4499_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_4499_end_mask_0 = const()[name = tensor<string, []>("op_4499_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4499_cast = slice_by_index(begin = var_4499_begin_0, end = var_4499_end_0, end_mask = var_4499_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4499_cast")];
            tensor<int32, [4]> var_4503_begin_0 = const()[name = tensor<string, []>("op_4503_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4503_end_0 = const()[name = tensor<string, []>("op_4503_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_4503_end_mask_0 = const()[name = tensor<string, []>("op_4503_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4503_cast = slice_by_index(begin = var_4503_begin_0, end = var_4503_end_0, end_mask = var_4503_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4503_cast")];
            tensor<int32, [4]> var_4507_begin_0 = const()[name = tensor<string, []>("op_4507_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4507_end_0 = const()[name = tensor<string, []>("op_4507_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_4507_end_mask_0 = const()[name = tensor<string, []>("op_4507_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4507_cast = slice_by_index(begin = var_4507_begin_0, end = var_4507_end_0, end_mask = var_4507_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4507_cast")];
            tensor<int32, [4]> var_4511_begin_0 = const()[name = tensor<string, []>("op_4511_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4511_end_0 = const()[name = tensor<string, []>("op_4511_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_4511_end_mask_0 = const()[name = tensor<string, []>("op_4511_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4511_cast = slice_by_index(begin = var_4511_begin_0, end = var_4511_end_0, end_mask = var_4511_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4511_cast")];
            tensor<int32, [4]> var_4515_begin_0 = const()[name = tensor<string, []>("op_4515_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4515_end_0 = const()[name = tensor<string, []>("op_4515_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_4515_end_mask_0 = const()[name = tensor<string, []>("op_4515_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4515_cast = slice_by_index(begin = var_4515_begin_0, end = var_4515_end_0, end_mask = var_4515_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4515_cast")];
            tensor<int32, [4]> var_4519_begin_0 = const()[name = tensor<string, []>("op_4519_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4519_end_0 = const()[name = tensor<string, []>("op_4519_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_4519_end_mask_0 = const()[name = tensor<string, []>("op_4519_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4519_cast = slice_by_index(begin = var_4519_begin_0, end = var_4519_end_0, end_mask = var_4519_end_mask_0, x = q_19_cast)[name = tensor<string, []>("op_4519_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_4526_begin_0 = const()[name = tensor<string, []>("op_4526_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4526_end_0 = const()[name = tensor<string, []>("op_4526_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_4526_end_mask_0 = const()[name = tensor<string, []>("op_4526_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, 64]> var_4526_cast = slice_by_index(begin = var_4526_begin_0, end = var_4526_end_0, end_mask = var_4526_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4526_cast")];
            tensor<int32, [4]> var_4530_begin_0 = const()[name = tensor<string, []>("op_4530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_4530_end_0 = const()[name = tensor<string, []>("op_4530_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_4530_end_mask_0 = const()[name = tensor<string, []>("op_4530_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4530_cast = slice_by_index(begin = var_4530_begin_0, end = var_4530_end_0, end_mask = var_4530_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4530_cast")];
            tensor<int32, [4]> var_4534_begin_0 = const()[name = tensor<string, []>("op_4534_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_4534_end_0 = const()[name = tensor<string, []>("op_4534_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_4534_end_mask_0 = const()[name = tensor<string, []>("op_4534_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4534_cast = slice_by_index(begin = var_4534_begin_0, end = var_4534_end_0, end_mask = var_4534_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4534_cast")];
            tensor<int32, [4]> var_4538_begin_0 = const()[name = tensor<string, []>("op_4538_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_4538_end_0 = const()[name = tensor<string, []>("op_4538_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_4538_end_mask_0 = const()[name = tensor<string, []>("op_4538_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4538_cast = slice_by_index(begin = var_4538_begin_0, end = var_4538_end_0, end_mask = var_4538_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4538_cast")];
            tensor<int32, [4]> var_4542_begin_0 = const()[name = tensor<string, []>("op_4542_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_4542_end_0 = const()[name = tensor<string, []>("op_4542_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_4542_end_mask_0 = const()[name = tensor<string, []>("op_4542_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4542_cast = slice_by_index(begin = var_4542_begin_0, end = var_4542_end_0, end_mask = var_4542_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4542_cast")];
            tensor<int32, [4]> var_4546_begin_0 = const()[name = tensor<string, []>("op_4546_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_4546_end_0 = const()[name = tensor<string, []>("op_4546_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_4546_end_mask_0 = const()[name = tensor<string, []>("op_4546_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4546_cast = slice_by_index(begin = var_4546_begin_0, end = var_4546_end_0, end_mask = var_4546_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4546_cast")];
            tensor<int32, [4]> var_4550_begin_0 = const()[name = tensor<string, []>("op_4550_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_4550_end_0 = const()[name = tensor<string, []>("op_4550_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_4550_end_mask_0 = const()[name = tensor<string, []>("op_4550_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4550_cast = slice_by_index(begin = var_4550_begin_0, end = var_4550_end_0, end_mask = var_4550_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4550_cast")];
            tensor<int32, [4]> var_4554_begin_0 = const()[name = tensor<string, []>("op_4554_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_4554_end_0 = const()[name = tensor<string, []>("op_4554_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_4554_end_mask_0 = const()[name = tensor<string, []>("op_4554_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4554_cast = slice_by_index(begin = var_4554_begin_0, end = var_4554_end_0, end_mask = var_4554_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4554_cast")];
            tensor<int32, [4]> var_4558_begin_0 = const()[name = tensor<string, []>("op_4558_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_4558_end_0 = const()[name = tensor<string, []>("op_4558_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_4558_end_mask_0 = const()[name = tensor<string, []>("op_4558_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4558_cast = slice_by_index(begin = var_4558_begin_0, end = var_4558_end_0, end_mask = var_4558_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4558_cast")];
            tensor<int32, [4]> var_4562_begin_0 = const()[name = tensor<string, []>("op_4562_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_4562_end_0 = const()[name = tensor<string, []>("op_4562_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_4562_end_mask_0 = const()[name = tensor<string, []>("op_4562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4562_cast = slice_by_index(begin = var_4562_begin_0, end = var_4562_end_0, end_mask = var_4562_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4562_cast")];
            tensor<int32, [4]> var_4566_begin_0 = const()[name = tensor<string, []>("op_4566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_4566_end_0 = const()[name = tensor<string, []>("op_4566_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_4566_end_mask_0 = const()[name = tensor<string, []>("op_4566_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4566_cast = slice_by_index(begin = var_4566_begin_0, end = var_4566_end_0, end_mask = var_4566_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4566_cast")];
            tensor<int32, [4]> var_4570_begin_0 = const()[name = tensor<string, []>("op_4570_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_4570_end_0 = const()[name = tensor<string, []>("op_4570_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_4570_end_mask_0 = const()[name = tensor<string, []>("op_4570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4570_cast = slice_by_index(begin = var_4570_begin_0, end = var_4570_end_0, end_mask = var_4570_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4570_cast")];
            tensor<int32, [4]> var_4574_begin_0 = const()[name = tensor<string, []>("op_4574_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_4574_end_0 = const()[name = tensor<string, []>("op_4574_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_4574_end_mask_0 = const()[name = tensor<string, []>("op_4574_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4574_cast = slice_by_index(begin = var_4574_begin_0, end = var_4574_end_0, end_mask = var_4574_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4574_cast")];
            tensor<int32, [4]> var_4578_begin_0 = const()[name = tensor<string, []>("op_4578_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_4578_end_0 = const()[name = tensor<string, []>("op_4578_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_4578_end_mask_0 = const()[name = tensor<string, []>("op_4578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4578_cast = slice_by_index(begin = var_4578_begin_0, end = var_4578_end_0, end_mask = var_4578_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4578_cast")];
            tensor<int32, [4]> var_4582_begin_0 = const()[name = tensor<string, []>("op_4582_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_4582_end_0 = const()[name = tensor<string, []>("op_4582_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_4582_end_mask_0 = const()[name = tensor<string, []>("op_4582_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4582_cast = slice_by_index(begin = var_4582_begin_0, end = var_4582_end_0, end_mask = var_4582_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4582_cast")];
            tensor<int32, [4]> var_4586_begin_0 = const()[name = tensor<string, []>("op_4586_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_4586_end_0 = const()[name = tensor<string, []>("op_4586_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_4586_end_mask_0 = const()[name = tensor<string, []>("op_4586_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4586_cast = slice_by_index(begin = var_4586_begin_0, end = var_4586_end_0, end_mask = var_4586_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4586_cast")];
            tensor<int32, [4]> var_4590_begin_0 = const()[name = tensor<string, []>("op_4590_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_4590_end_0 = const()[name = tensor<string, []>("op_4590_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_4590_end_mask_0 = const()[name = tensor<string, []>("op_4590_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4590_cast = slice_by_index(begin = var_4590_begin_0, end = var_4590_end_0, end_mask = var_4590_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4590_cast")];
            tensor<int32, [4]> var_4594_begin_0 = const()[name = tensor<string, []>("op_4594_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_4594_end_0 = const()[name = tensor<string, []>("op_4594_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_4594_end_mask_0 = const()[name = tensor<string, []>("op_4594_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4594_cast = slice_by_index(begin = var_4594_begin_0, end = var_4594_end_0, end_mask = var_4594_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4594_cast")];
            tensor<int32, [4]> var_4598_begin_0 = const()[name = tensor<string, []>("op_4598_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_4598_end_0 = const()[name = tensor<string, []>("op_4598_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_4598_end_mask_0 = const()[name = tensor<string, []>("op_4598_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4598_cast = slice_by_index(begin = var_4598_begin_0, end = var_4598_end_0, end_mask = var_4598_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4598_cast")];
            tensor<int32, [4]> var_4602_begin_0 = const()[name = tensor<string, []>("op_4602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_4602_end_0 = const()[name = tensor<string, []>("op_4602_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_4602_end_mask_0 = const()[name = tensor<string, []>("op_4602_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_4602_cast = slice_by_index(begin = var_4602_begin_0, end = var_4602_end_0, end_mask = var_4602_end_mask_0, x = transpose_22)[name = tensor<string, []>("op_4602_cast")];
            tensor<int32, [4]> var_4604_begin_0 = const()[name = tensor<string, []>("op_4604_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4604_end_0 = const()[name = tensor<string, []>("op_4604_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_4604_end_mask_0 = const()[name = tensor<string, []>("op_4604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4604_cast = slice_by_index(begin = var_4604_begin_0, end = var_4604_end_0, end_mask = var_4604_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4604_cast")];
            tensor<int32, [4]> var_4608_begin_0 = const()[name = tensor<string, []>("op_4608_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4608_end_0 = const()[name = tensor<string, []>("op_4608_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_4608_end_mask_0 = const()[name = tensor<string, []>("op_4608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4608_cast = slice_by_index(begin = var_4608_begin_0, end = var_4608_end_0, end_mask = var_4608_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4608_cast")];
            tensor<int32, [4]> var_4612_begin_0 = const()[name = tensor<string, []>("op_4612_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_4612_end_0 = const()[name = tensor<string, []>("op_4612_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_4612_end_mask_0 = const()[name = tensor<string, []>("op_4612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4612_cast = slice_by_index(begin = var_4612_begin_0, end = var_4612_end_0, end_mask = var_4612_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4612_cast")];
            tensor<int32, [4]> var_4616_begin_0 = const()[name = tensor<string, []>("op_4616_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_4616_end_0 = const()[name = tensor<string, []>("op_4616_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_4616_end_mask_0 = const()[name = tensor<string, []>("op_4616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4616_cast = slice_by_index(begin = var_4616_begin_0, end = var_4616_end_0, end_mask = var_4616_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4616_cast")];
            tensor<int32, [4]> var_4620_begin_0 = const()[name = tensor<string, []>("op_4620_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_4620_end_0 = const()[name = tensor<string, []>("op_4620_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_4620_end_mask_0 = const()[name = tensor<string, []>("op_4620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4620_cast = slice_by_index(begin = var_4620_begin_0, end = var_4620_end_0, end_mask = var_4620_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4620_cast")];
            tensor<int32, [4]> var_4624_begin_0 = const()[name = tensor<string, []>("op_4624_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_4624_end_0 = const()[name = tensor<string, []>("op_4624_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_4624_end_mask_0 = const()[name = tensor<string, []>("op_4624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4624_cast = slice_by_index(begin = var_4624_begin_0, end = var_4624_end_0, end_mask = var_4624_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4624_cast")];
            tensor<int32, [4]> var_4628_begin_0 = const()[name = tensor<string, []>("op_4628_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_4628_end_0 = const()[name = tensor<string, []>("op_4628_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_4628_end_mask_0 = const()[name = tensor<string, []>("op_4628_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4628_cast = slice_by_index(begin = var_4628_begin_0, end = var_4628_end_0, end_mask = var_4628_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4628_cast")];
            tensor<int32, [4]> var_4632_begin_0 = const()[name = tensor<string, []>("op_4632_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_4632_end_0 = const()[name = tensor<string, []>("op_4632_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_4632_end_mask_0 = const()[name = tensor<string, []>("op_4632_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4632_cast = slice_by_index(begin = var_4632_begin_0, end = var_4632_end_0, end_mask = var_4632_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4632_cast")];
            tensor<int32, [4]> var_4636_begin_0 = const()[name = tensor<string, []>("op_4636_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_4636_end_0 = const()[name = tensor<string, []>("op_4636_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_4636_end_mask_0 = const()[name = tensor<string, []>("op_4636_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4636_cast = slice_by_index(begin = var_4636_begin_0, end = var_4636_end_0, end_mask = var_4636_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4636_cast")];
            tensor<int32, [4]> var_4640_begin_0 = const()[name = tensor<string, []>("op_4640_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_4640_end_0 = const()[name = tensor<string, []>("op_4640_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_4640_end_mask_0 = const()[name = tensor<string, []>("op_4640_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4640_cast = slice_by_index(begin = var_4640_begin_0, end = var_4640_end_0, end_mask = var_4640_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4640_cast")];
            tensor<int32, [4]> var_4644_begin_0 = const()[name = tensor<string, []>("op_4644_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_4644_end_0 = const()[name = tensor<string, []>("op_4644_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_4644_end_mask_0 = const()[name = tensor<string, []>("op_4644_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4644_cast = slice_by_index(begin = var_4644_begin_0, end = var_4644_end_0, end_mask = var_4644_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4644_cast")];
            tensor<int32, [4]> var_4648_begin_0 = const()[name = tensor<string, []>("op_4648_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_4648_end_0 = const()[name = tensor<string, []>("op_4648_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_4648_end_mask_0 = const()[name = tensor<string, []>("op_4648_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4648_cast = slice_by_index(begin = var_4648_begin_0, end = var_4648_end_0, end_mask = var_4648_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4648_cast")];
            tensor<int32, [4]> var_4652_begin_0 = const()[name = tensor<string, []>("op_4652_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_4652_end_0 = const()[name = tensor<string, []>("op_4652_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_4652_end_mask_0 = const()[name = tensor<string, []>("op_4652_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4652_cast = slice_by_index(begin = var_4652_begin_0, end = var_4652_end_0, end_mask = var_4652_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4652_cast")];
            tensor<int32, [4]> var_4656_begin_0 = const()[name = tensor<string, []>("op_4656_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_4656_end_0 = const()[name = tensor<string, []>("op_4656_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_4656_end_mask_0 = const()[name = tensor<string, []>("op_4656_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4656_cast = slice_by_index(begin = var_4656_begin_0, end = var_4656_end_0, end_mask = var_4656_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4656_cast")];
            tensor<int32, [4]> var_4660_begin_0 = const()[name = tensor<string, []>("op_4660_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_4660_end_0 = const()[name = tensor<string, []>("op_4660_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_4660_end_mask_0 = const()[name = tensor<string, []>("op_4660_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4660_cast = slice_by_index(begin = var_4660_begin_0, end = var_4660_end_0, end_mask = var_4660_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4660_cast")];
            tensor<int32, [4]> var_4664_begin_0 = const()[name = tensor<string, []>("op_4664_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_4664_end_0 = const()[name = tensor<string, []>("op_4664_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_4664_end_mask_0 = const()[name = tensor<string, []>("op_4664_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4664_cast = slice_by_index(begin = var_4664_begin_0, end = var_4664_end_0, end_mask = var_4664_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4664_cast")];
            tensor<int32, [4]> var_4668_begin_0 = const()[name = tensor<string, []>("op_4668_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_4668_end_0 = const()[name = tensor<string, []>("op_4668_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_4668_end_mask_0 = const()[name = tensor<string, []>("op_4668_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4668_cast = slice_by_index(begin = var_4668_begin_0, end = var_4668_end_0, end_mask = var_4668_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4668_cast")];
            tensor<int32, [4]> var_4672_begin_0 = const()[name = tensor<string, []>("op_4672_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_4672_end_0 = const()[name = tensor<string, []>("op_4672_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_4672_end_mask_0 = const()[name = tensor<string, []>("op_4672_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4672_cast = slice_by_index(begin = var_4672_begin_0, end = var_4672_end_0, end_mask = var_4672_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4672_cast")];
            tensor<int32, [4]> var_4676_begin_0 = const()[name = tensor<string, []>("op_4676_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_4676_end_0 = const()[name = tensor<string, []>("op_4676_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_4676_end_mask_0 = const()[name = tensor<string, []>("op_4676_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4676_cast = slice_by_index(begin = var_4676_begin_0, end = var_4676_end_0, end_mask = var_4676_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4676_cast")];
            tensor<int32, [4]> var_4680_begin_0 = const()[name = tensor<string, []>("op_4680_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_4680_end_0 = const()[name = tensor<string, []>("op_4680_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_4680_end_mask_0 = const()[name = tensor<string, []>("op_4680_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_4680_cast = slice_by_index(begin = var_4680_begin_0, end = var_4680_end_0, end_mask = var_4680_end_mask_0, x = v_19_cast)[name = tensor<string, []>("op_4680_cast")];
            tensor<string, []> var_4684_equation_0 = const()[name = tensor<string, []>("op_4684_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4684_cast = einsum(equation = var_4684_equation_0, values = (var_4526_cast, var_4443_cast))[name = tensor<string, []>("op_4684_cast")];
            tensor<fp16, []> var_4685_to_fp16 = const()[name = tensor<string, []>("op_4685_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_41_cast = mul(x = var_4684_cast, y = var_4685_to_fp16)[name = tensor<string, []>("aw_41_cast")];
            tensor<string, []> var_4688_equation_0 = const()[name = tensor<string, []>("op_4688_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4688_cast = einsum(equation = var_4688_equation_0, values = (var_4530_cast, var_4447_cast))[name = tensor<string, []>("op_4688_cast")];
            tensor<fp16, []> var_4689_to_fp16 = const()[name = tensor<string, []>("op_4689_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_43_cast = mul(x = var_4688_cast, y = var_4689_to_fp16)[name = tensor<string, []>("aw_43_cast")];
            tensor<string, []> var_4692_equation_0 = const()[name = tensor<string, []>("op_4692_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4692_cast = einsum(equation = var_4692_equation_0, values = (var_4534_cast, var_4451_cast))[name = tensor<string, []>("op_4692_cast")];
            tensor<fp16, []> var_4693_to_fp16 = const()[name = tensor<string, []>("op_4693_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_45_cast = mul(x = var_4692_cast, y = var_4693_to_fp16)[name = tensor<string, []>("aw_45_cast")];
            tensor<string, []> var_4696_equation_0 = const()[name = tensor<string, []>("op_4696_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4696_cast = einsum(equation = var_4696_equation_0, values = (var_4538_cast, var_4455_cast))[name = tensor<string, []>("op_4696_cast")];
            tensor<fp16, []> var_4697_to_fp16 = const()[name = tensor<string, []>("op_4697_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_47_cast = mul(x = var_4696_cast, y = var_4697_to_fp16)[name = tensor<string, []>("aw_47_cast")];
            tensor<string, []> var_4700_equation_0 = const()[name = tensor<string, []>("op_4700_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4700_cast = einsum(equation = var_4700_equation_0, values = (var_4542_cast, var_4459_cast))[name = tensor<string, []>("op_4700_cast")];
            tensor<fp16, []> var_4701_to_fp16 = const()[name = tensor<string, []>("op_4701_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_49_cast = mul(x = var_4700_cast, y = var_4701_to_fp16)[name = tensor<string, []>("aw_49_cast")];
            tensor<string, []> var_4704_equation_0 = const()[name = tensor<string, []>("op_4704_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4704_cast = einsum(equation = var_4704_equation_0, values = (var_4546_cast, var_4463_cast))[name = tensor<string, []>("op_4704_cast")];
            tensor<fp16, []> var_4705_to_fp16 = const()[name = tensor<string, []>("op_4705_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_51_cast = mul(x = var_4704_cast, y = var_4705_to_fp16)[name = tensor<string, []>("aw_51_cast")];
            tensor<string, []> var_4708_equation_0 = const()[name = tensor<string, []>("op_4708_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4708_cast = einsum(equation = var_4708_equation_0, values = (var_4550_cast, var_4467_cast))[name = tensor<string, []>("op_4708_cast")];
            tensor<fp16, []> var_4709_to_fp16 = const()[name = tensor<string, []>("op_4709_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_53_cast = mul(x = var_4708_cast, y = var_4709_to_fp16)[name = tensor<string, []>("aw_53_cast")];
            tensor<string, []> var_4712_equation_0 = const()[name = tensor<string, []>("op_4712_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4712_cast = einsum(equation = var_4712_equation_0, values = (var_4554_cast, var_4471_cast))[name = tensor<string, []>("op_4712_cast")];
            tensor<fp16, []> var_4713_to_fp16 = const()[name = tensor<string, []>("op_4713_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_55_cast = mul(x = var_4712_cast, y = var_4713_to_fp16)[name = tensor<string, []>("aw_55_cast")];
            tensor<string, []> var_4716_equation_0 = const()[name = tensor<string, []>("op_4716_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4716_cast = einsum(equation = var_4716_equation_0, values = (var_4558_cast, var_4475_cast))[name = tensor<string, []>("op_4716_cast")];
            tensor<fp16, []> var_4717_to_fp16 = const()[name = tensor<string, []>("op_4717_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_57_cast = mul(x = var_4716_cast, y = var_4717_to_fp16)[name = tensor<string, []>("aw_57_cast")];
            tensor<string, []> var_4720_equation_0 = const()[name = tensor<string, []>("op_4720_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4720_cast = einsum(equation = var_4720_equation_0, values = (var_4562_cast, var_4479_cast))[name = tensor<string, []>("op_4720_cast")];
            tensor<fp16, []> var_4721_to_fp16 = const()[name = tensor<string, []>("op_4721_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_59_cast = mul(x = var_4720_cast, y = var_4721_to_fp16)[name = tensor<string, []>("aw_59_cast")];
            tensor<string, []> var_4724_equation_0 = const()[name = tensor<string, []>("op_4724_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4724_cast = einsum(equation = var_4724_equation_0, values = (var_4566_cast, var_4483_cast))[name = tensor<string, []>("op_4724_cast")];
            tensor<fp16, []> var_4725_to_fp16 = const()[name = tensor<string, []>("op_4725_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_61_cast = mul(x = var_4724_cast, y = var_4725_to_fp16)[name = tensor<string, []>("aw_61_cast")];
            tensor<string, []> var_4728_equation_0 = const()[name = tensor<string, []>("op_4728_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4728_cast = einsum(equation = var_4728_equation_0, values = (var_4570_cast, var_4487_cast))[name = tensor<string, []>("op_4728_cast")];
            tensor<fp16, []> var_4729_to_fp16 = const()[name = tensor<string, []>("op_4729_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_63_cast = mul(x = var_4728_cast, y = var_4729_to_fp16)[name = tensor<string, []>("aw_63_cast")];
            tensor<string, []> var_4732_equation_0 = const()[name = tensor<string, []>("op_4732_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4732_cast = einsum(equation = var_4732_equation_0, values = (var_4574_cast, var_4491_cast))[name = tensor<string, []>("op_4732_cast")];
            tensor<fp16, []> var_4733_to_fp16 = const()[name = tensor<string, []>("op_4733_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_65_cast = mul(x = var_4732_cast, y = var_4733_to_fp16)[name = tensor<string, []>("aw_65_cast")];
            tensor<string, []> var_4736_equation_0 = const()[name = tensor<string, []>("op_4736_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4736_cast = einsum(equation = var_4736_equation_0, values = (var_4578_cast, var_4495_cast))[name = tensor<string, []>("op_4736_cast")];
            tensor<fp16, []> var_4737_to_fp16 = const()[name = tensor<string, []>("op_4737_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_67_cast = mul(x = var_4736_cast, y = var_4737_to_fp16)[name = tensor<string, []>("aw_67_cast")];
            tensor<string, []> var_4740_equation_0 = const()[name = tensor<string, []>("op_4740_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4740_cast = einsum(equation = var_4740_equation_0, values = (var_4582_cast, var_4499_cast))[name = tensor<string, []>("op_4740_cast")];
            tensor<fp16, []> var_4741_to_fp16 = const()[name = tensor<string, []>("op_4741_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_69_cast = mul(x = var_4740_cast, y = var_4741_to_fp16)[name = tensor<string, []>("aw_69_cast")];
            tensor<string, []> var_4744_equation_0 = const()[name = tensor<string, []>("op_4744_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4744_cast = einsum(equation = var_4744_equation_0, values = (var_4586_cast, var_4503_cast))[name = tensor<string, []>("op_4744_cast")];
            tensor<fp16, []> var_4745_to_fp16 = const()[name = tensor<string, []>("op_4745_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_71_cast = mul(x = var_4744_cast, y = var_4745_to_fp16)[name = tensor<string, []>("aw_71_cast")];
            tensor<string, []> var_4748_equation_0 = const()[name = tensor<string, []>("op_4748_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4748_cast = einsum(equation = var_4748_equation_0, values = (var_4590_cast, var_4507_cast))[name = tensor<string, []>("op_4748_cast")];
            tensor<fp16, []> var_4749_to_fp16 = const()[name = tensor<string, []>("op_4749_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_73_cast = mul(x = var_4748_cast, y = var_4749_to_fp16)[name = tensor<string, []>("aw_73_cast")];
            tensor<string, []> var_4752_equation_0 = const()[name = tensor<string, []>("op_4752_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4752_cast = einsum(equation = var_4752_equation_0, values = (var_4594_cast, var_4511_cast))[name = tensor<string, []>("op_4752_cast")];
            tensor<fp16, []> var_4753_to_fp16 = const()[name = tensor<string, []>("op_4753_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_75_cast = mul(x = var_4752_cast, y = var_4753_to_fp16)[name = tensor<string, []>("aw_75_cast")];
            tensor<string, []> var_4756_equation_0 = const()[name = tensor<string, []>("op_4756_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4756_cast = einsum(equation = var_4756_equation_0, values = (var_4598_cast, var_4515_cast))[name = tensor<string, []>("op_4756_cast")];
            tensor<fp16, []> var_4757_to_fp16 = const()[name = tensor<string, []>("op_4757_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_77_cast = mul(x = var_4756_cast, y = var_4757_to_fp16)[name = tensor<string, []>("aw_77_cast")];
            tensor<string, []> var_4760_equation_0 = const()[name = tensor<string, []>("op_4760_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_4760_cast = einsum(equation = var_4760_equation_0, values = (var_4602_cast, var_4519_cast))[name = tensor<string, []>("op_4760_cast")];
            tensor<fp16, []> var_4761_to_fp16 = const()[name = tensor<string, []>("op_4761_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_79_cast = mul(x = var_4760_cast, y = var_4761_to_fp16)[name = tensor<string, []>("aw_79_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4763_cast = softmax(axis = var_3883, x = aw_41_cast)[name = tensor<string, []>("op_4763_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4764_cast = softmax(axis = var_3883, x = aw_43_cast)[name = tensor<string, []>("op_4764_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4765_cast = softmax(axis = var_3883, x = aw_45_cast)[name = tensor<string, []>("op_4765_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4766_cast = softmax(axis = var_3883, x = aw_47_cast)[name = tensor<string, []>("op_4766_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4767_cast = softmax(axis = var_3883, x = aw_49_cast)[name = tensor<string, []>("op_4767_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4768_cast = softmax(axis = var_3883, x = aw_51_cast)[name = tensor<string, []>("op_4768_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4769_cast = softmax(axis = var_3883, x = aw_53_cast)[name = tensor<string, []>("op_4769_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4770_cast = softmax(axis = var_3883, x = aw_55_cast)[name = tensor<string, []>("op_4770_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4771_cast = softmax(axis = var_3883, x = aw_57_cast)[name = tensor<string, []>("op_4771_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4772_cast = softmax(axis = var_3883, x = aw_59_cast)[name = tensor<string, []>("op_4772_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4773_cast = softmax(axis = var_3883, x = aw_61_cast)[name = tensor<string, []>("op_4773_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4774_cast = softmax(axis = var_3883, x = aw_63_cast)[name = tensor<string, []>("op_4774_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4775_cast = softmax(axis = var_3883, x = aw_65_cast)[name = tensor<string, []>("op_4775_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4776_cast = softmax(axis = var_3883, x = aw_67_cast)[name = tensor<string, []>("op_4776_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4777_cast = softmax(axis = var_3883, x = aw_69_cast)[name = tensor<string, []>("op_4777_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4778_cast = softmax(axis = var_3883, x = aw_71_cast)[name = tensor<string, []>("op_4778_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4779_cast = softmax(axis = var_3883, x = aw_73_cast)[name = tensor<string, []>("op_4779_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4780_cast = softmax(axis = var_3883, x = aw_75_cast)[name = tensor<string, []>("op_4780_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4781_cast = softmax(axis = var_3883, x = aw_77_cast)[name = tensor<string, []>("op_4781_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_4782_cast = softmax(axis = var_3883, x = aw_79_cast)[name = tensor<string, []>("op_4782_cast")];
            tensor<string, []> var_4784_equation_0 = const()[name = tensor<string, []>("op_4784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4784_cast = einsum(equation = var_4784_equation_0, values = (var_4604_cast, var_4763_cast))[name = tensor<string, []>("op_4784_cast")];
            tensor<string, []> var_4786_equation_0 = const()[name = tensor<string, []>("op_4786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4786_cast = einsum(equation = var_4786_equation_0, values = (var_4608_cast, var_4764_cast))[name = tensor<string, []>("op_4786_cast")];
            tensor<string, []> var_4788_equation_0 = const()[name = tensor<string, []>("op_4788_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4788_cast = einsum(equation = var_4788_equation_0, values = (var_4612_cast, var_4765_cast))[name = tensor<string, []>("op_4788_cast")];
            tensor<string, []> var_4790_equation_0 = const()[name = tensor<string, []>("op_4790_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4790_cast = einsum(equation = var_4790_equation_0, values = (var_4616_cast, var_4766_cast))[name = tensor<string, []>("op_4790_cast")];
            tensor<string, []> var_4792_equation_0 = const()[name = tensor<string, []>("op_4792_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4792_cast = einsum(equation = var_4792_equation_0, values = (var_4620_cast, var_4767_cast))[name = tensor<string, []>("op_4792_cast")];
            tensor<string, []> var_4794_equation_0 = const()[name = tensor<string, []>("op_4794_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4794_cast = einsum(equation = var_4794_equation_0, values = (var_4624_cast, var_4768_cast))[name = tensor<string, []>("op_4794_cast")];
            tensor<string, []> var_4796_equation_0 = const()[name = tensor<string, []>("op_4796_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4796_cast = einsum(equation = var_4796_equation_0, values = (var_4628_cast, var_4769_cast))[name = tensor<string, []>("op_4796_cast")];
            tensor<string, []> var_4798_equation_0 = const()[name = tensor<string, []>("op_4798_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4798_cast = einsum(equation = var_4798_equation_0, values = (var_4632_cast, var_4770_cast))[name = tensor<string, []>("op_4798_cast")];
            tensor<string, []> var_4800_equation_0 = const()[name = tensor<string, []>("op_4800_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4800_cast = einsum(equation = var_4800_equation_0, values = (var_4636_cast, var_4771_cast))[name = tensor<string, []>("op_4800_cast")];
            tensor<string, []> var_4802_equation_0 = const()[name = tensor<string, []>("op_4802_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4802_cast = einsum(equation = var_4802_equation_0, values = (var_4640_cast, var_4772_cast))[name = tensor<string, []>("op_4802_cast")];
            tensor<string, []> var_4804_equation_0 = const()[name = tensor<string, []>("op_4804_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4804_cast = einsum(equation = var_4804_equation_0, values = (var_4644_cast, var_4773_cast))[name = tensor<string, []>("op_4804_cast")];
            tensor<string, []> var_4806_equation_0 = const()[name = tensor<string, []>("op_4806_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4806_cast = einsum(equation = var_4806_equation_0, values = (var_4648_cast, var_4774_cast))[name = tensor<string, []>("op_4806_cast")];
            tensor<string, []> var_4808_equation_0 = const()[name = tensor<string, []>("op_4808_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4808_cast = einsum(equation = var_4808_equation_0, values = (var_4652_cast, var_4775_cast))[name = tensor<string, []>("op_4808_cast")];
            tensor<string, []> var_4810_equation_0 = const()[name = tensor<string, []>("op_4810_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4810_cast = einsum(equation = var_4810_equation_0, values = (var_4656_cast, var_4776_cast))[name = tensor<string, []>("op_4810_cast")];
            tensor<string, []> var_4812_equation_0 = const()[name = tensor<string, []>("op_4812_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4812_cast = einsum(equation = var_4812_equation_0, values = (var_4660_cast, var_4777_cast))[name = tensor<string, []>("op_4812_cast")];
            tensor<string, []> var_4814_equation_0 = const()[name = tensor<string, []>("op_4814_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4814_cast = einsum(equation = var_4814_equation_0, values = (var_4664_cast, var_4778_cast))[name = tensor<string, []>("op_4814_cast")];
            tensor<string, []> var_4816_equation_0 = const()[name = tensor<string, []>("op_4816_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4816_cast = einsum(equation = var_4816_equation_0, values = (var_4668_cast, var_4779_cast))[name = tensor<string, []>("op_4816_cast")];
            tensor<string, []> var_4818_equation_0 = const()[name = tensor<string, []>("op_4818_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4818_cast = einsum(equation = var_4818_equation_0, values = (var_4672_cast, var_4780_cast))[name = tensor<string, []>("op_4818_cast")];
            tensor<string, []> var_4820_equation_0 = const()[name = tensor<string, []>("op_4820_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4820_cast = einsum(equation = var_4820_equation_0, values = (var_4676_cast, var_4781_cast))[name = tensor<string, []>("op_4820_cast")];
            tensor<string, []> var_4822_equation_0 = const()[name = tensor<string, []>("op_4822_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_4822_cast = einsum(equation = var_4822_equation_0, values = (var_4680_cast, var_4782_cast))[name = tensor<string, []>("op_4822_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_3883, interleave = input_135_interleave_0, values = (var_4784_cast, var_4786_cast, var_4788_cast, var_4790_cast, var_4792_cast, var_4794_cast, var_4796_cast, var_4798_cast, var_4800_cast, var_4802_cast, var_4804_cast, var_4806_cast, var_4808_cast, var_4810_cast, var_4812_cast, var_4814_cast, var_4816_cast, var_4818_cast, var_4820_cast, var_4822_cast))[name = tensor<string, []>("input_135_cast")];
            tensor<int32, [2]> var_4828 = const()[name = tensor<string, []>("op_4828"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4830 = const()[name = tensor<string, []>("op_4830"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4832_pad_type_0 = const()[name = tensor<string, []>("op_4832_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4832_pad_0 = const()[name = tensor<string, []>("op_4832_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21987968))), lut = tensor<fp16, [4]>([-0x1.56cp-5, -0x1.994p-7, 0x1.96p-7, 0x1.564p-5]), 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, []>(22397632)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4832_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_4830, groups = var_3883, pad = var_4832_pad_0, pad_type = var_4832_pad_type_0, strides = var_4828, 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_4832_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_29_cast = add(x = var_4832_cast, y = inputs_27_cast)[name = tensor<string, []>("inputs_29_cast")];
            tensor<int32, [1]> var_4836 = const()[name = tensor<string, []>("op_4836"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_29_cast = reduce_mean(axes = var_4836, keep_dims = var_3878, 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_4840 = const()[name = tensor<string, []>("op_4840"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_4841_cast = reduce_mean(axes = var_4840, keep_dims = var_3878, x = zero_mean_sq_29_cast)[name = tensor<string, []>("op_4841_cast")];
            tensor<fp16, []> var_4842_to_fp16 = const()[name = tensor<string, []>("op_4842_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_4843_cast = add(x = var_4841_cast, y = var_4842_to_fp16)[name = tensor<string, []>("op_4843_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_4843_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_4847_to_fp16 = const()[name = tensor<string, []>("op_4847_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22400256)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4848_cast = add(x = out_29_cast, y = var_4847_to_fp16)[name = tensor<string, []>("op_4848_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4850_to_fp16 = const()[name = tensor<string, []>("op_4850_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22402880)))];
            tensor<fp16, [2, 1280, 1, 256]> input_137_cast = mul(x = var_4848_cast, y = var_4850_to_fp16)[name = tensor<string, []>("input_137_cast")];
            tensor<int32, [2]> var_4858 = const()[name = tensor<string, []>("op_4858"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4860 = const()[name = tensor<string, []>("op_4860"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4862_pad_type_0 = const()[name = tensor<string, []>("op_4862_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4862_pad_0 = const()[name = tensor<string, []>("op_4862_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, [3276800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22405504))), lut = tensor<fp16, [4]>([-0x1.6acp-4, -0x1.aacp-6, 0x1.a94p-6, 0x1.6a8p-4]), 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, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25682368))), lut = tensor<fp16, [4]>([-0x1.7e8p-3, -0x1.6ap-4, -0x1.638p-6, 0x1.2acp-6]), 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_4862_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_4860, groups = var_3883, pad = var_4862_pad_0, pad_type = var_4862_pad_type_0, strides = var_4858, 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_4862_cast")];
            tensor<int32, [2]> var_4863_split_sizes_0 = const()[name = tensor<string, []>("op_4863_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_4863_axis_0 = const()[name = tensor<string, []>("op_4863_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_4863_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_4863_cast_1 = split(axis = var_4863_axis_0, split_sizes = var_4863_split_sizes_0, x = var_4862_cast)[name = tensor<string, []>("op_4863_cast")];
            tensor<string, []> var_4865_mode_0 = const()[name = tensor<string, []>("op_4865_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_4865_cast = gelu(mode = var_4865_mode_0, x = var_4863_cast_1)[name = tensor<string, []>("op_4865_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_139_cast = mul(x = var_4863_cast_0, y = var_4865_cast)[name = tensor<string, []>("input_139_cast")];
            tensor<int32, [2]> var_4869 = const()[name = tensor<string, []>("op_4869"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4871 = const()[name = tensor<string, []>("op_4871"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_4873_pad_type_0 = const()[name = tensor<string, []>("op_4873_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_4873_pad_0 = const()[name = tensor<string, []>("op_4873_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, [1638400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25684992))), lut = tensor<fp16, [4]>([-0x1.5e4p-4, -0x1.a28p-6, 0x1.a04p-6, 0x1.5dcp-4]), 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, []>(27323456)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4873_cast = conv(bias = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_4871, groups = var_3883, pad = var_4873_pad_0, pad_type = var_4873_pad_type_0, strides = var_4869, weight = down_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_139_cast)[name = tensor<string, []>("op_4873_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_87_cast = add(x = var_4873_cast, y = inputs_29_cast)[name = tensor<string, []>("hidden_states_87_cast")];
            tensor<int32, [4]> var_4875 = const()[name = tensor<string, []>("op_4875"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_141_cast = reshape(shape = var_4875, x = hidden_states_87_cast)[name = tensor<string, []>("input_141_cast")];
            tensor<int32, [2]> var_4879 = const()[name = tensor<string, []>("op_4879"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4881 = const()[name = tensor<string, []>("op_4881"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27326080))), lut = tensor<fp16, [4]>([-0x1.084p-4, -0x1.3c4p-6, 0x1.3d4p-6, 0x1.084p-4]), 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, []>(27735744)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_89_cast = conv(bias = down_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_4881, groups = var_3883, pad = hidden_states_89_pad_0, pad_type = hidden_states_89_pad_type_0, strides = var_4879, 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, []>(27738368)))];
            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, []>(27740992)))];
            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_4896 = const()[name = tensor<string, []>("op_4896"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4898 = const()[name = tensor<string, []>("op_4898"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27743616))), lut = tensor<fp16, [4]>([-0x1.188p-4, -0x1.4f8p-6, 0x1.33p-6, 0x1.0e8p-4]), 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, []>(31430080)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_91_cast = conv(bias = down_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_4898, groups = var_3883, pad = hidden_states_91_pad_0, pad_type = hidden_states_91_pad_type_0, strides = var_4896, 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_4904 = const()[name = tensor<string, []>("op_4904"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4906 = const()[name = tensor<string, []>("op_4906"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31432704))), lut = tensor<fp16, [4]>([-0x1.d78p+0, -0x1.004p-8, 0x1.a1cp-9, 0x1.658p-5]), 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, []>(31842368)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_11_cast = conv(bias = down_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_4906, groups = var_3883, pad = temb_11_pad_0, pad_type = temb_11_pad_type_0, strides = var_4904, 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, []>(31844992)))];
            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, []>(31847616)))];
            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_4916 = const()[name = tensor<string, []>("op_4916"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4918 = const()[name = tensor<string, []>("op_4918"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31850240))), lut = tensor<fp16, [4]>([-0x1.fb8p-5, -0x1.2acp-6, 0x1.264p-6, 0x1.f8p-5]), 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, []>(35536704)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_93_cast = conv(bias = down_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_4918, groups = var_3883, pad = hidden_states_93_pad_0, pad_type = hidden_states_93_pad_type_0, strides = var_4916, 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, []>(35539328)))];
            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, []>(35541952)))];
            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_4938 = const()[name = tensor<string, []>("op_4938"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4940 = const()[name = tensor<string, []>("op_4940"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35544576))), lut = tensor<fp16, [4]>([-0x1.944p-4, -0x1.e84p-6, 0x1.d94p-6, 0x1.8ep-4]), 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, []>(35954240)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_97_cast = conv(bias = down_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_4940, groups = var_3883, pad = hidden_states_97_pad_0, pad_type = hidden_states_97_pad_type_0, strides = var_4938, 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_4945 = const()[name = tensor<string, []>("op_4945"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_31_cast = reshape(shape = var_4945, x = hidden_states_97_cast)[name = tensor<string, []>("inputs_31_cast")];
            tensor<int32, [1]> var_4955 = const()[name = tensor<string, []>("op_4955"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_31_cast = reduce_mean(axes = var_4955, keep_dims = var_3878, 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_4959 = const()[name = tensor<string, []>("op_4959"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_4960_cast = reduce_mean(axes = var_4959, keep_dims = var_3878, x = zero_mean_sq_31_cast)[name = tensor<string, []>("op_4960_cast")];
            tensor<fp16, []> var_4961_to_fp16 = const()[name = tensor<string, []>("op_4961_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_4962_cast = add(x = var_4960_cast, y = var_4961_to_fp16)[name = tensor<string, []>("op_4962_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_4962_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_4966_to_fp16 = const()[name = tensor<string, []>("op_4966_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35956864)))];
            tensor<fp16, [2, 1280, 1, 256]> var_4967_cast = add(x = out_31_cast, y = var_4966_to_fp16)[name = tensor<string, []>("op_4967_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_4969_to_fp16 = const()[name = tensor<string, []>("op_4969_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35959488)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_99_cast = mul(x = var_4967_cast, y = var_4969_to_fp16)[name = tensor<string, []>("hidden_states_99_cast")];
            tensor<int32, [2]> var_4976 = const()[name = tensor<string, []>("op_4976"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4978 = const()[name = tensor<string, []>("op_4978"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35962112))), lut = tensor<fp16, [4]>([-0x1.888p-4, -0x1.d6cp-6, 0x1.ce8p-6, 0x1.85cp-4]), 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_4978, groups = var_3883, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = var_4976, 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_4982 = const()[name = tensor<string, []>("op_4982"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4984 = const()[name = tensor<string, []>("op_4984"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36371776))), lut = tensor<fp16, [4]>([-0x1.894p-4, -0x1.d68p-6, 0x1.d3p-6, 0x1.88p-4]), 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_4984, groups = var_3883, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = var_4982, 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_4988 = const()[name = tensor<string, []>("op_4988"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_4990 = const()[name = tensor<string, []>("op_4990"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36781440))), lut = tensor<fp16, [4]>([-0x1.344p-4, -0x1.72p-6, 0x1.6c8p-6, 0x1.334p-4]), 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_4990, groups = var_3883, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = var_4988, 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_4994_begin_0 = const()[name = tensor<string, []>("op_4994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_4994_end_0 = const()[name = tensor<string, []>("op_4994_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            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, 64, 1, 256]> var_4994_cast = slice_by_index(begin = var_4994_begin_0, end = var_4994_end_0, end_mask = var_4994_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_4994_cast")];
            tensor<int32, [4]> var_4998_begin_0 = const()[name = tensor<string, []>("op_4998_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_4998_end_0 = const()[name = tensor<string, []>("op_4998_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_4998_end_mask_0 = const()[name = tensor<string, []>("op_4998_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_4998_cast = slice_by_index(begin = var_4998_begin_0, end = var_4998_end_0, end_mask = var_4998_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_4998_cast")];
            tensor<int32, [4]> var_5002_begin_0 = const()[name = tensor<string, []>("op_5002_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5002_end_0 = const()[name = tensor<string, []>("op_5002_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_5002_end_mask_0 = const()[name = tensor<string, []>("op_5002_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5002_cast = slice_by_index(begin = var_5002_begin_0, end = var_5002_end_0, end_mask = var_5002_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5002_cast")];
            tensor<int32, [4]> var_5006_begin_0 = const()[name = tensor<string, []>("op_5006_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5006_end_0 = const()[name = tensor<string, []>("op_5006_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5006_end_mask_0 = const()[name = tensor<string, []>("op_5006_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5006_cast = slice_by_index(begin = var_5006_begin_0, end = var_5006_end_0, end_mask = var_5006_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5006_cast")];
            tensor<int32, [4]> var_5010_begin_0 = const()[name = tensor<string, []>("op_5010_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5010_end_0 = const()[name = tensor<string, []>("op_5010_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5010_end_mask_0 = const()[name = tensor<string, []>("op_5010_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5010_cast = slice_by_index(begin = var_5010_begin_0, end = var_5010_end_0, end_mask = var_5010_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5010_cast")];
            tensor<int32, [4]> var_5014_begin_0 = const()[name = tensor<string, []>("op_5014_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5014_end_0 = const()[name = tensor<string, []>("op_5014_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_5014_end_mask_0 = const()[name = tensor<string, []>("op_5014_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5014_cast = slice_by_index(begin = var_5014_begin_0, end = var_5014_end_0, end_mask = var_5014_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5014_cast")];
            tensor<int32, [4]> var_5018_begin_0 = const()[name = tensor<string, []>("op_5018_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5018_end_0 = const()[name = tensor<string, []>("op_5018_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_5018_end_mask_0 = const()[name = tensor<string, []>("op_5018_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5018_cast = slice_by_index(begin = var_5018_begin_0, end = var_5018_end_0, end_mask = var_5018_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5018_cast")];
            tensor<int32, [4]> var_5022_begin_0 = const()[name = tensor<string, []>("op_5022_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5022_end_0 = const()[name = tensor<string, []>("op_5022_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_5022_end_mask_0 = const()[name = tensor<string, []>("op_5022_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5022_cast = slice_by_index(begin = var_5022_begin_0, end = var_5022_end_0, end_mask = var_5022_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5022_cast")];
            tensor<int32, [4]> var_5026_begin_0 = const()[name = tensor<string, []>("op_5026_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5026_end_0 = const()[name = tensor<string, []>("op_5026_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_5026_end_mask_0 = const()[name = tensor<string, []>("op_5026_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5026_cast = slice_by_index(begin = var_5026_begin_0, end = var_5026_end_0, end_mask = var_5026_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5026_cast")];
            tensor<int32, [4]> var_5030_begin_0 = const()[name = tensor<string, []>("op_5030_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5030_end_0 = const()[name = tensor<string, []>("op_5030_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5030_end_mask_0 = const()[name = tensor<string, []>("op_5030_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5030_cast = slice_by_index(begin = var_5030_begin_0, end = var_5030_end_0, end_mask = var_5030_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5030_cast")];
            tensor<int32, [4]> var_5034_begin_0 = const()[name = tensor<string, []>("op_5034_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5034_end_0 = const()[name = tensor<string, []>("op_5034_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_5034_end_mask_0 = const()[name = tensor<string, []>("op_5034_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5034_cast = slice_by_index(begin = var_5034_begin_0, end = var_5034_end_0, end_mask = var_5034_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5034_cast")];
            tensor<int32, [4]> var_5038_begin_0 = const()[name = tensor<string, []>("op_5038_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5038_end_0 = const()[name = tensor<string, []>("op_5038_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_5038_end_mask_0 = const()[name = tensor<string, []>("op_5038_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5038_cast = slice_by_index(begin = var_5038_begin_0, end = var_5038_end_0, end_mask = var_5038_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5038_cast")];
            tensor<int32, [4]> var_5042_begin_0 = const()[name = tensor<string, []>("op_5042_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5042_end_0 = const()[name = tensor<string, []>("op_5042_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_5042_end_mask_0 = const()[name = tensor<string, []>("op_5042_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5042_cast = slice_by_index(begin = var_5042_begin_0, end = var_5042_end_0, end_mask = var_5042_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5042_cast")];
            tensor<int32, [4]> var_5046_begin_0 = const()[name = tensor<string, []>("op_5046_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5046_end_0 = const()[name = tensor<string, []>("op_5046_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_5046_end_mask_0 = const()[name = tensor<string, []>("op_5046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5046_cast = slice_by_index(begin = var_5046_begin_0, end = var_5046_end_0, end_mask = var_5046_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5046_cast")];
            tensor<int32, [4]> var_5050_begin_0 = const()[name = tensor<string, []>("op_5050_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5050_end_0 = const()[name = tensor<string, []>("op_5050_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5050_end_mask_0 = const()[name = tensor<string, []>("op_5050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5050_cast = slice_by_index(begin = var_5050_begin_0, end = var_5050_end_0, end_mask = var_5050_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5050_cast")];
            tensor<int32, [4]> var_5054_begin_0 = const()[name = tensor<string, []>("op_5054_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5054_end_0 = const()[name = tensor<string, []>("op_5054_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_5054_end_mask_0 = const()[name = tensor<string, []>("op_5054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5054_cast = slice_by_index(begin = var_5054_begin_0, end = var_5054_end_0, end_mask = var_5054_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5054_cast")];
            tensor<int32, [4]> var_5058_begin_0 = const()[name = tensor<string, []>("op_5058_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5058_end_0 = const()[name = tensor<string, []>("op_5058_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_5058_end_mask_0 = const()[name = tensor<string, []>("op_5058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5058_cast = slice_by_index(begin = var_5058_begin_0, end = var_5058_end_0, end_mask = var_5058_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5058_cast")];
            tensor<int32, [4]> var_5062_begin_0 = const()[name = tensor<string, []>("op_5062_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5062_end_0 = const()[name = tensor<string, []>("op_5062_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_5062_end_mask_0 = const()[name = tensor<string, []>("op_5062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5062_cast = slice_by_index(begin = var_5062_begin_0, end = var_5062_end_0, end_mask = var_5062_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5062_cast")];
            tensor<int32, [4]> var_5066_begin_0 = const()[name = tensor<string, []>("op_5066_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5066_end_0 = const()[name = tensor<string, []>("op_5066_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_5066_end_mask_0 = const()[name = tensor<string, []>("op_5066_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5066_cast = slice_by_index(begin = var_5066_begin_0, end = var_5066_end_0, end_mask = var_5066_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5066_cast")];
            tensor<int32, [4]> var_5070_begin_0 = const()[name = tensor<string, []>("op_5070_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5070_end_0 = const()[name = tensor<string, []>("op_5070_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5070_end_mask_0 = const()[name = tensor<string, []>("op_5070_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5070_cast = slice_by_index(begin = var_5070_begin_0, end = var_5070_end_0, end_mask = var_5070_end_mask_0, x = q_21_cast)[name = tensor<string, []>("op_5070_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_5077_begin_0 = const()[name = tensor<string, []>("op_5077_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5077_end_0 = const()[name = tensor<string, []>("op_5077_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_5077_end_mask_0 = const()[name = tensor<string, []>("op_5077_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, 64]> var_5077_cast = slice_by_index(begin = var_5077_begin_0, end = var_5077_end_0, end_mask = var_5077_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5077_cast")];
            tensor<int32, [4]> var_5081_begin_0 = const()[name = tensor<string, []>("op_5081_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_5081_end_0 = const()[name = tensor<string, []>("op_5081_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_5081_end_mask_0 = const()[name = tensor<string, []>("op_5081_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5081_cast = slice_by_index(begin = var_5081_begin_0, end = var_5081_end_0, end_mask = var_5081_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5081_cast")];
            tensor<int32, [4]> var_5085_begin_0 = const()[name = tensor<string, []>("op_5085_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_5085_end_0 = const()[name = tensor<string, []>("op_5085_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_5085_end_mask_0 = const()[name = tensor<string, []>("op_5085_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5085_cast = slice_by_index(begin = var_5085_begin_0, end = var_5085_end_0, end_mask = var_5085_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5085_cast")];
            tensor<int32, [4]> var_5089_begin_0 = const()[name = tensor<string, []>("op_5089_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_5089_end_0 = const()[name = tensor<string, []>("op_5089_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5089_end_mask_0 = const()[name = tensor<string, []>("op_5089_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5089_cast = slice_by_index(begin = var_5089_begin_0, end = var_5089_end_0, end_mask = var_5089_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5089_cast")];
            tensor<int32, [4]> var_5093_begin_0 = const()[name = tensor<string, []>("op_5093_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_5093_end_0 = const()[name = tensor<string, []>("op_5093_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_5093_end_mask_0 = const()[name = tensor<string, []>("op_5093_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5093_cast = slice_by_index(begin = var_5093_begin_0, end = var_5093_end_0, end_mask = var_5093_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5093_cast")];
            tensor<int32, [4]> var_5097_begin_0 = const()[name = tensor<string, []>("op_5097_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5097_end_0 = const()[name = tensor<string, []>("op_5097_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_5097_end_mask_0 = const()[name = tensor<string, []>("op_5097_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5097_cast = slice_by_index(begin = var_5097_begin_0, end = var_5097_end_0, end_mask = var_5097_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5097_cast")];
            tensor<int32, [4]> var_5101_begin_0 = const()[name = tensor<string, []>("op_5101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_5101_end_0 = const()[name = tensor<string, []>("op_5101_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_5101_end_mask_0 = const()[name = tensor<string, []>("op_5101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5101_cast = slice_by_index(begin = var_5101_begin_0, end = var_5101_end_0, end_mask = var_5101_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5101_cast")];
            tensor<int32, [4]> var_5105_begin_0 = const()[name = tensor<string, []>("op_5105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_5105_end_0 = const()[name = tensor<string, []>("op_5105_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_5105_end_mask_0 = const()[name = tensor<string, []>("op_5105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5105_cast = slice_by_index(begin = var_5105_begin_0, end = var_5105_end_0, end_mask = var_5105_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5105_cast")];
            tensor<int32, [4]> var_5109_begin_0 = const()[name = tensor<string, []>("op_5109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_5109_end_0 = const()[name = tensor<string, []>("op_5109_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_5109_end_mask_0 = const()[name = tensor<string, []>("op_5109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5109_cast = slice_by_index(begin = var_5109_begin_0, end = var_5109_end_0, end_mask = var_5109_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5109_cast")];
            tensor<int32, [4]> var_5113_begin_0 = const()[name = tensor<string, []>("op_5113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_5113_end_0 = const()[name = tensor<string, []>("op_5113_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_5113_end_mask_0 = const()[name = tensor<string, []>("op_5113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5113_cast = slice_by_index(begin = var_5113_begin_0, end = var_5113_end_0, end_mask = var_5113_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5113_cast")];
            tensor<int32, [4]> var_5117_begin_0 = const()[name = tensor<string, []>("op_5117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5117_end_0 = const()[name = tensor<string, []>("op_5117_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_5117_end_mask_0 = const()[name = tensor<string, []>("op_5117_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5117_cast = slice_by_index(begin = var_5117_begin_0, end = var_5117_end_0, end_mask = var_5117_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5117_cast")];
            tensor<int32, [4]> var_5121_begin_0 = const()[name = tensor<string, []>("op_5121_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_5121_end_0 = const()[name = tensor<string, []>("op_5121_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_5121_end_mask_0 = const()[name = tensor<string, []>("op_5121_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5121_cast = slice_by_index(begin = var_5121_begin_0, end = var_5121_end_0, end_mask = var_5121_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5121_cast")];
            tensor<int32, [4]> var_5125_begin_0 = const()[name = tensor<string, []>("op_5125_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_5125_end_0 = const()[name = tensor<string, []>("op_5125_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_5125_end_mask_0 = const()[name = tensor<string, []>("op_5125_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5125_cast = slice_by_index(begin = var_5125_begin_0, end = var_5125_end_0, end_mask = var_5125_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5125_cast")];
            tensor<int32, [4]> var_5129_begin_0 = const()[name = tensor<string, []>("op_5129_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_5129_end_0 = const()[name = tensor<string, []>("op_5129_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_5129_end_mask_0 = const()[name = tensor<string, []>("op_5129_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5129_cast = slice_by_index(begin = var_5129_begin_0, end = var_5129_end_0, end_mask = var_5129_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5129_cast")];
            tensor<int32, [4]> var_5133_begin_0 = const()[name = tensor<string, []>("op_5133_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_5133_end_0 = const()[name = tensor<string, []>("op_5133_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_5133_end_mask_0 = const()[name = tensor<string, []>("op_5133_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5133_cast = slice_by_index(begin = var_5133_begin_0, end = var_5133_end_0, end_mask = var_5133_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5133_cast")];
            tensor<int32, [4]> var_5137_begin_0 = const()[name = tensor<string, []>("op_5137_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5137_end_0 = const()[name = tensor<string, []>("op_5137_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_5137_end_mask_0 = const()[name = tensor<string, []>("op_5137_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5137_cast = slice_by_index(begin = var_5137_begin_0, end = var_5137_end_0, end_mask = var_5137_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5137_cast")];
            tensor<int32, [4]> var_5141_begin_0 = const()[name = tensor<string, []>("op_5141_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_5141_end_0 = const()[name = tensor<string, []>("op_5141_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_5141_end_mask_0 = const()[name = tensor<string, []>("op_5141_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5141_cast = slice_by_index(begin = var_5141_begin_0, end = var_5141_end_0, end_mask = var_5141_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5141_cast")];
            tensor<int32, [4]> var_5145_begin_0 = const()[name = tensor<string, []>("op_5145_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_5145_end_0 = const()[name = tensor<string, []>("op_5145_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_5145_end_mask_0 = const()[name = tensor<string, []>("op_5145_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5145_cast = slice_by_index(begin = var_5145_begin_0, end = var_5145_end_0, end_mask = var_5145_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5145_cast")];
            tensor<int32, [4]> var_5149_begin_0 = const()[name = tensor<string, []>("op_5149_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_5149_end_0 = const()[name = tensor<string, []>("op_5149_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_5149_end_mask_0 = const()[name = tensor<string, []>("op_5149_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5149_cast = slice_by_index(begin = var_5149_begin_0, end = var_5149_end_0, end_mask = var_5149_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5149_cast")];
            tensor<int32, [4]> var_5153_begin_0 = const()[name = tensor<string, []>("op_5153_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_5153_end_0 = const()[name = tensor<string, []>("op_5153_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_5153_end_mask_0 = const()[name = tensor<string, []>("op_5153_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_5153_cast = slice_by_index(begin = var_5153_begin_0, end = var_5153_end_0, end_mask = var_5153_end_mask_0, x = transpose_21)[name = tensor<string, []>("op_5153_cast")];
            tensor<int32, [4]> var_5155_begin_0 = const()[name = tensor<string, []>("op_5155_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5155_end_0 = const()[name = tensor<string, []>("op_5155_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_5155_end_mask_0 = const()[name = tensor<string, []>("op_5155_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5155_cast = slice_by_index(begin = var_5155_begin_0, end = var_5155_end_0, end_mask = var_5155_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5155_cast")];
            tensor<int32, [4]> var_5159_begin_0 = const()[name = tensor<string, []>("op_5159_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5159_end_0 = const()[name = tensor<string, []>("op_5159_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_5159_end_mask_0 = const()[name = tensor<string, []>("op_5159_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5159_cast = slice_by_index(begin = var_5159_begin_0, end = var_5159_end_0, end_mask = var_5159_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5159_cast")];
            tensor<int32, [4]> var_5163_begin_0 = const()[name = tensor<string, []>("op_5163_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5163_end_0 = const()[name = tensor<string, []>("op_5163_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_5163_end_mask_0 = const()[name = tensor<string, []>("op_5163_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5163_cast = slice_by_index(begin = var_5163_begin_0, end = var_5163_end_0, end_mask = var_5163_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5163_cast")];
            tensor<int32, [4]> var_5167_begin_0 = const()[name = tensor<string, []>("op_5167_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5167_end_0 = const()[name = tensor<string, []>("op_5167_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5167_end_mask_0 = const()[name = tensor<string, []>("op_5167_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5167_cast = slice_by_index(begin = var_5167_begin_0, end = var_5167_end_0, end_mask = var_5167_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5167_cast")];
            tensor<int32, [4]> var_5171_begin_0 = const()[name = tensor<string, []>("op_5171_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5171_end_0 = const()[name = tensor<string, []>("op_5171_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5171_end_mask_0 = const()[name = tensor<string, []>("op_5171_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5171_cast = slice_by_index(begin = var_5171_begin_0, end = var_5171_end_0, end_mask = var_5171_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5171_cast")];
            tensor<int32, [4]> var_5175_begin_0 = const()[name = tensor<string, []>("op_5175_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5175_end_0 = const()[name = tensor<string, []>("op_5175_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_5175_end_mask_0 = const()[name = tensor<string, []>("op_5175_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5175_cast = slice_by_index(begin = var_5175_begin_0, end = var_5175_end_0, end_mask = var_5175_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5175_cast")];
            tensor<int32, [4]> var_5179_begin_0 = const()[name = tensor<string, []>("op_5179_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5179_end_0 = const()[name = tensor<string, []>("op_5179_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_5179_end_mask_0 = const()[name = tensor<string, []>("op_5179_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5179_cast = slice_by_index(begin = var_5179_begin_0, end = var_5179_end_0, end_mask = var_5179_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5179_cast")];
            tensor<int32, [4]> var_5183_begin_0 = const()[name = tensor<string, []>("op_5183_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5183_end_0 = const()[name = tensor<string, []>("op_5183_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_5183_end_mask_0 = const()[name = tensor<string, []>("op_5183_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5183_cast = slice_by_index(begin = var_5183_begin_0, end = var_5183_end_0, end_mask = var_5183_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5183_cast")];
            tensor<int32, [4]> var_5187_begin_0 = const()[name = tensor<string, []>("op_5187_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5187_end_0 = const()[name = tensor<string, []>("op_5187_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_5187_end_mask_0 = const()[name = tensor<string, []>("op_5187_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5187_cast = slice_by_index(begin = var_5187_begin_0, end = var_5187_end_0, end_mask = var_5187_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5187_cast")];
            tensor<int32, [4]> var_5191_begin_0 = const()[name = tensor<string, []>("op_5191_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5191_end_0 = const()[name = tensor<string, []>("op_5191_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5191_end_mask_0 = const()[name = tensor<string, []>("op_5191_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5191_cast = slice_by_index(begin = var_5191_begin_0, end = var_5191_end_0, end_mask = var_5191_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5191_cast")];
            tensor<int32, [4]> var_5195_begin_0 = const()[name = tensor<string, []>("op_5195_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5195_end_0 = const()[name = tensor<string, []>("op_5195_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_5195_end_mask_0 = const()[name = tensor<string, []>("op_5195_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5195_cast = slice_by_index(begin = var_5195_begin_0, end = var_5195_end_0, end_mask = var_5195_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5195_cast")];
            tensor<int32, [4]> var_5199_begin_0 = const()[name = tensor<string, []>("op_5199_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5199_end_0 = const()[name = tensor<string, []>("op_5199_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_5199_end_mask_0 = const()[name = tensor<string, []>("op_5199_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5199_cast = slice_by_index(begin = var_5199_begin_0, end = var_5199_end_0, end_mask = var_5199_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5199_cast")];
            tensor<int32, [4]> var_5203_begin_0 = const()[name = tensor<string, []>("op_5203_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5203_end_0 = const()[name = tensor<string, []>("op_5203_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_5203_end_mask_0 = const()[name = tensor<string, []>("op_5203_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5203_cast = slice_by_index(begin = var_5203_begin_0, end = var_5203_end_0, end_mask = var_5203_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5203_cast")];
            tensor<int32, [4]> var_5207_begin_0 = const()[name = tensor<string, []>("op_5207_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5207_end_0 = const()[name = tensor<string, []>("op_5207_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_5207_end_mask_0 = const()[name = tensor<string, []>("op_5207_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5207_cast = slice_by_index(begin = var_5207_begin_0, end = var_5207_end_0, end_mask = var_5207_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5207_cast")];
            tensor<int32, [4]> var_5211_begin_0 = const()[name = tensor<string, []>("op_5211_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5211_end_0 = const()[name = tensor<string, []>("op_5211_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5211_end_mask_0 = const()[name = tensor<string, []>("op_5211_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5211_cast = slice_by_index(begin = var_5211_begin_0, end = var_5211_end_0, end_mask = var_5211_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5211_cast")];
            tensor<int32, [4]> var_5215_begin_0 = const()[name = tensor<string, []>("op_5215_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5215_end_0 = const()[name = tensor<string, []>("op_5215_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_5215_end_mask_0 = const()[name = tensor<string, []>("op_5215_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5215_cast = slice_by_index(begin = var_5215_begin_0, end = var_5215_end_0, end_mask = var_5215_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5215_cast")];
            tensor<int32, [4]> var_5219_begin_0 = const()[name = tensor<string, []>("op_5219_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5219_end_0 = const()[name = tensor<string, []>("op_5219_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_5219_end_mask_0 = const()[name = tensor<string, []>("op_5219_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5219_cast = slice_by_index(begin = var_5219_begin_0, end = var_5219_end_0, end_mask = var_5219_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5219_cast")];
            tensor<int32, [4]> var_5223_begin_0 = const()[name = tensor<string, []>("op_5223_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5223_end_0 = const()[name = tensor<string, []>("op_5223_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_5223_end_mask_0 = const()[name = tensor<string, []>("op_5223_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5223_cast = slice_by_index(begin = var_5223_begin_0, end = var_5223_end_0, end_mask = var_5223_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5223_cast")];
            tensor<int32, [4]> var_5227_begin_0 = const()[name = tensor<string, []>("op_5227_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5227_end_0 = const()[name = tensor<string, []>("op_5227_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_5227_end_mask_0 = const()[name = tensor<string, []>("op_5227_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5227_cast = slice_by_index(begin = var_5227_begin_0, end = var_5227_end_0, end_mask = var_5227_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5227_cast")];
            tensor<int32, [4]> var_5231_begin_0 = const()[name = tensor<string, []>("op_5231_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5231_end_0 = const()[name = tensor<string, []>("op_5231_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5231_end_mask_0 = const()[name = tensor<string, []>("op_5231_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5231_cast = slice_by_index(begin = var_5231_begin_0, end = var_5231_end_0, end_mask = var_5231_end_mask_0, x = v_21_cast)[name = tensor<string, []>("op_5231_cast")];
            tensor<string, []> var_5235_equation_0 = const()[name = tensor<string, []>("op_5235_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5235_cast = einsum(equation = var_5235_equation_0, values = (var_5077_cast, var_4994_cast))[name = tensor<string, []>("op_5235_cast")];
            tensor<fp16, []> var_5236_to_fp16 = const()[name = tensor<string, []>("op_5236_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_81_cast = mul(x = var_5235_cast, y = var_5236_to_fp16)[name = tensor<string, []>("aw_81_cast")];
            tensor<string, []> var_5239_equation_0 = const()[name = tensor<string, []>("op_5239_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5239_cast = einsum(equation = var_5239_equation_0, values = (var_5081_cast, var_4998_cast))[name = tensor<string, []>("op_5239_cast")];
            tensor<fp16, []> var_5240_to_fp16 = const()[name = tensor<string, []>("op_5240_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_83_cast = mul(x = var_5239_cast, y = var_5240_to_fp16)[name = tensor<string, []>("aw_83_cast")];
            tensor<string, []> var_5243_equation_0 = const()[name = tensor<string, []>("op_5243_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5243_cast = einsum(equation = var_5243_equation_0, values = (var_5085_cast, var_5002_cast))[name = tensor<string, []>("op_5243_cast")];
            tensor<fp16, []> var_5244_to_fp16 = const()[name = tensor<string, []>("op_5244_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_85_cast = mul(x = var_5243_cast, y = var_5244_to_fp16)[name = tensor<string, []>("aw_85_cast")];
            tensor<string, []> var_5247_equation_0 = const()[name = tensor<string, []>("op_5247_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5247_cast = einsum(equation = var_5247_equation_0, values = (var_5089_cast, var_5006_cast))[name = tensor<string, []>("op_5247_cast")];
            tensor<fp16, []> var_5248_to_fp16 = const()[name = tensor<string, []>("op_5248_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_87_cast = mul(x = var_5247_cast, y = var_5248_to_fp16)[name = tensor<string, []>("aw_87_cast")];
            tensor<string, []> var_5251_equation_0 = const()[name = tensor<string, []>("op_5251_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5251_cast = einsum(equation = var_5251_equation_0, values = (var_5093_cast, var_5010_cast))[name = tensor<string, []>("op_5251_cast")];
            tensor<fp16, []> var_5252_to_fp16 = const()[name = tensor<string, []>("op_5252_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_89_cast = mul(x = var_5251_cast, y = var_5252_to_fp16)[name = tensor<string, []>("aw_89_cast")];
            tensor<string, []> var_5255_equation_0 = const()[name = tensor<string, []>("op_5255_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5255_cast = einsum(equation = var_5255_equation_0, values = (var_5097_cast, var_5014_cast))[name = tensor<string, []>("op_5255_cast")];
            tensor<fp16, []> var_5256_to_fp16 = const()[name = tensor<string, []>("op_5256_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_91_cast = mul(x = var_5255_cast, y = var_5256_to_fp16)[name = tensor<string, []>("aw_91_cast")];
            tensor<string, []> var_5259_equation_0 = const()[name = tensor<string, []>("op_5259_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5259_cast = einsum(equation = var_5259_equation_0, values = (var_5101_cast, var_5018_cast))[name = tensor<string, []>("op_5259_cast")];
            tensor<fp16, []> var_5260_to_fp16 = const()[name = tensor<string, []>("op_5260_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_93_cast = mul(x = var_5259_cast, y = var_5260_to_fp16)[name = tensor<string, []>("aw_93_cast")];
            tensor<string, []> var_5263_equation_0 = const()[name = tensor<string, []>("op_5263_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5263_cast = einsum(equation = var_5263_equation_0, values = (var_5105_cast, var_5022_cast))[name = tensor<string, []>("op_5263_cast")];
            tensor<fp16, []> var_5264_to_fp16 = const()[name = tensor<string, []>("op_5264_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_95_cast = mul(x = var_5263_cast, y = var_5264_to_fp16)[name = tensor<string, []>("aw_95_cast")];
            tensor<string, []> var_5267_equation_0 = const()[name = tensor<string, []>("op_5267_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5267_cast = einsum(equation = var_5267_equation_0, values = (var_5109_cast, var_5026_cast))[name = tensor<string, []>("op_5267_cast")];
            tensor<fp16, []> var_5268_to_fp16 = const()[name = tensor<string, []>("op_5268_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_97_cast = mul(x = var_5267_cast, y = var_5268_to_fp16)[name = tensor<string, []>("aw_97_cast")];
            tensor<string, []> var_5271_equation_0 = const()[name = tensor<string, []>("op_5271_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5271_cast = einsum(equation = var_5271_equation_0, values = (var_5113_cast, var_5030_cast))[name = tensor<string, []>("op_5271_cast")];
            tensor<fp16, []> var_5272_to_fp16 = const()[name = tensor<string, []>("op_5272_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_99_cast = mul(x = var_5271_cast, y = var_5272_to_fp16)[name = tensor<string, []>("aw_99_cast")];
            tensor<string, []> var_5275_equation_0 = const()[name = tensor<string, []>("op_5275_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5275_cast = einsum(equation = var_5275_equation_0, values = (var_5117_cast, var_5034_cast))[name = tensor<string, []>("op_5275_cast")];
            tensor<fp16, []> var_5276_to_fp16 = const()[name = tensor<string, []>("op_5276_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_101_cast = mul(x = var_5275_cast, y = var_5276_to_fp16)[name = tensor<string, []>("aw_101_cast")];
            tensor<string, []> var_5279_equation_0 = const()[name = tensor<string, []>("op_5279_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5279_cast = einsum(equation = var_5279_equation_0, values = (var_5121_cast, var_5038_cast))[name = tensor<string, []>("op_5279_cast")];
            tensor<fp16, []> var_5280_to_fp16 = const()[name = tensor<string, []>("op_5280_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_103_cast = mul(x = var_5279_cast, y = var_5280_to_fp16)[name = tensor<string, []>("aw_103_cast")];
            tensor<string, []> var_5283_equation_0 = const()[name = tensor<string, []>("op_5283_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5283_cast = einsum(equation = var_5283_equation_0, values = (var_5125_cast, var_5042_cast))[name = tensor<string, []>("op_5283_cast")];
            tensor<fp16, []> var_5284_to_fp16 = const()[name = tensor<string, []>("op_5284_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_105_cast = mul(x = var_5283_cast, y = var_5284_to_fp16)[name = tensor<string, []>("aw_105_cast")];
            tensor<string, []> var_5287_equation_0 = const()[name = tensor<string, []>("op_5287_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5287_cast = einsum(equation = var_5287_equation_0, values = (var_5129_cast, var_5046_cast))[name = tensor<string, []>("op_5287_cast")];
            tensor<fp16, []> var_5288_to_fp16 = const()[name = tensor<string, []>("op_5288_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_107_cast = mul(x = var_5287_cast, y = var_5288_to_fp16)[name = tensor<string, []>("aw_107_cast")];
            tensor<string, []> var_5291_equation_0 = const()[name = tensor<string, []>("op_5291_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5291_cast = einsum(equation = var_5291_equation_0, values = (var_5133_cast, var_5050_cast))[name = tensor<string, []>("op_5291_cast")];
            tensor<fp16, []> var_5292_to_fp16 = const()[name = tensor<string, []>("op_5292_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_109_cast = mul(x = var_5291_cast, y = var_5292_to_fp16)[name = tensor<string, []>("aw_109_cast")];
            tensor<string, []> var_5295_equation_0 = const()[name = tensor<string, []>("op_5295_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5295_cast = einsum(equation = var_5295_equation_0, values = (var_5137_cast, var_5054_cast))[name = tensor<string, []>("op_5295_cast")];
            tensor<fp16, []> var_5296_to_fp16 = const()[name = tensor<string, []>("op_5296_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_111_cast = mul(x = var_5295_cast, y = var_5296_to_fp16)[name = tensor<string, []>("aw_111_cast")];
            tensor<string, []> var_5299_equation_0 = const()[name = tensor<string, []>("op_5299_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5299_cast = einsum(equation = var_5299_equation_0, values = (var_5141_cast, var_5058_cast))[name = tensor<string, []>("op_5299_cast")];
            tensor<fp16, []> var_5300_to_fp16 = const()[name = tensor<string, []>("op_5300_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_113_cast = mul(x = var_5299_cast, y = var_5300_to_fp16)[name = tensor<string, []>("aw_113_cast")];
            tensor<string, []> var_5303_equation_0 = const()[name = tensor<string, []>("op_5303_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5303_cast = einsum(equation = var_5303_equation_0, values = (var_5145_cast, var_5062_cast))[name = tensor<string, []>("op_5303_cast")];
            tensor<fp16, []> var_5304_to_fp16 = const()[name = tensor<string, []>("op_5304_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_115_cast = mul(x = var_5303_cast, y = var_5304_to_fp16)[name = tensor<string, []>("aw_115_cast")];
            tensor<string, []> var_5307_equation_0 = const()[name = tensor<string, []>("op_5307_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5307_cast = einsum(equation = var_5307_equation_0, values = (var_5149_cast, var_5066_cast))[name = tensor<string, []>("op_5307_cast")];
            tensor<fp16, []> var_5308_to_fp16 = const()[name = tensor<string, []>("op_5308_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_117_cast = mul(x = var_5307_cast, y = var_5308_to_fp16)[name = tensor<string, []>("aw_117_cast")];
            tensor<string, []> var_5311_equation_0 = const()[name = tensor<string, []>("op_5311_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_5311_cast = einsum(equation = var_5311_equation_0, values = (var_5153_cast, var_5070_cast))[name = tensor<string, []>("op_5311_cast")];
            tensor<fp16, []> var_5312_to_fp16 = const()[name = tensor<string, []>("op_5312_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_119_cast = mul(x = var_5311_cast, y = var_5312_to_fp16)[name = tensor<string, []>("aw_119_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5314_cast = softmax(axis = var_3883, x = aw_81_cast)[name = tensor<string, []>("op_5314_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5315_cast = softmax(axis = var_3883, x = aw_83_cast)[name = tensor<string, []>("op_5315_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5316_cast = softmax(axis = var_3883, x = aw_85_cast)[name = tensor<string, []>("op_5316_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5317_cast = softmax(axis = var_3883, x = aw_87_cast)[name = tensor<string, []>("op_5317_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5318_cast = softmax(axis = var_3883, x = aw_89_cast)[name = tensor<string, []>("op_5318_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5319_cast = softmax(axis = var_3883, x = aw_91_cast)[name = tensor<string, []>("op_5319_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5320_cast = softmax(axis = var_3883, x = aw_93_cast)[name = tensor<string, []>("op_5320_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5321_cast = softmax(axis = var_3883, x = aw_95_cast)[name = tensor<string, []>("op_5321_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5322_cast = softmax(axis = var_3883, x = aw_97_cast)[name = tensor<string, []>("op_5322_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5323_cast = softmax(axis = var_3883, x = aw_99_cast)[name = tensor<string, []>("op_5323_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5324_cast = softmax(axis = var_3883, x = aw_101_cast)[name = tensor<string, []>("op_5324_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5325_cast = softmax(axis = var_3883, x = aw_103_cast)[name = tensor<string, []>("op_5325_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5326_cast = softmax(axis = var_3883, x = aw_105_cast)[name = tensor<string, []>("op_5326_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5327_cast = softmax(axis = var_3883, x = aw_107_cast)[name = tensor<string, []>("op_5327_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5328_cast = softmax(axis = var_3883, x = aw_109_cast)[name = tensor<string, []>("op_5328_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5329_cast = softmax(axis = var_3883, x = aw_111_cast)[name = tensor<string, []>("op_5329_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5330_cast = softmax(axis = var_3883, x = aw_113_cast)[name = tensor<string, []>("op_5330_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5331_cast = softmax(axis = var_3883, x = aw_115_cast)[name = tensor<string, []>("op_5331_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5332_cast = softmax(axis = var_3883, x = aw_117_cast)[name = tensor<string, []>("op_5332_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_5333_cast = softmax(axis = var_3883, x = aw_119_cast)[name = tensor<string, []>("op_5333_cast")];
            tensor<string, []> var_5335_equation_0 = const()[name = tensor<string, []>("op_5335_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5335_cast = einsum(equation = var_5335_equation_0, values = (var_5155_cast, var_5314_cast))[name = tensor<string, []>("op_5335_cast")];
            tensor<string, []> var_5337_equation_0 = const()[name = tensor<string, []>("op_5337_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5337_cast = einsum(equation = var_5337_equation_0, values = (var_5159_cast, var_5315_cast))[name = tensor<string, []>("op_5337_cast")];
            tensor<string, []> var_5339_equation_0 = const()[name = tensor<string, []>("op_5339_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5339_cast = einsum(equation = var_5339_equation_0, values = (var_5163_cast, var_5316_cast))[name = tensor<string, []>("op_5339_cast")];
            tensor<string, []> var_5341_equation_0 = const()[name = tensor<string, []>("op_5341_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5341_cast = einsum(equation = var_5341_equation_0, values = (var_5167_cast, var_5317_cast))[name = tensor<string, []>("op_5341_cast")];
            tensor<string, []> var_5343_equation_0 = const()[name = tensor<string, []>("op_5343_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5343_cast = einsum(equation = var_5343_equation_0, values = (var_5171_cast, var_5318_cast))[name = tensor<string, []>("op_5343_cast")];
            tensor<string, []> var_5345_equation_0 = const()[name = tensor<string, []>("op_5345_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5345_cast = einsum(equation = var_5345_equation_0, values = (var_5175_cast, var_5319_cast))[name = tensor<string, []>("op_5345_cast")];
            tensor<string, []> var_5347_equation_0 = const()[name = tensor<string, []>("op_5347_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5347_cast = einsum(equation = var_5347_equation_0, values = (var_5179_cast, var_5320_cast))[name = tensor<string, []>("op_5347_cast")];
            tensor<string, []> var_5349_equation_0 = const()[name = tensor<string, []>("op_5349_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5349_cast = einsum(equation = var_5349_equation_0, values = (var_5183_cast, var_5321_cast))[name = tensor<string, []>("op_5349_cast")];
            tensor<string, []> var_5351_equation_0 = const()[name = tensor<string, []>("op_5351_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5351_cast = einsum(equation = var_5351_equation_0, values = (var_5187_cast, var_5322_cast))[name = tensor<string, []>("op_5351_cast")];
            tensor<string, []> var_5353_equation_0 = const()[name = tensor<string, []>("op_5353_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5353_cast = einsum(equation = var_5353_equation_0, values = (var_5191_cast, var_5323_cast))[name = tensor<string, []>("op_5353_cast")];
            tensor<string, []> var_5355_equation_0 = const()[name = tensor<string, []>("op_5355_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5355_cast = einsum(equation = var_5355_equation_0, values = (var_5195_cast, var_5324_cast))[name = tensor<string, []>("op_5355_cast")];
            tensor<string, []> var_5357_equation_0 = const()[name = tensor<string, []>("op_5357_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5357_cast = einsum(equation = var_5357_equation_0, values = (var_5199_cast, var_5325_cast))[name = tensor<string, []>("op_5357_cast")];
            tensor<string, []> var_5359_equation_0 = const()[name = tensor<string, []>("op_5359_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5359_cast = einsum(equation = var_5359_equation_0, values = (var_5203_cast, var_5326_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, 64, 1, 256]> var_5361_cast = einsum(equation = var_5361_equation_0, values = (var_5207_cast, var_5327_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, 64, 1, 256]> var_5363_cast = einsum(equation = var_5363_equation_0, values = (var_5211_cast, var_5328_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, 64, 1, 256]> var_5365_cast = einsum(equation = var_5365_equation_0, values = (var_5215_cast, var_5329_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, 64, 1, 256]> var_5367_cast = einsum(equation = var_5367_equation_0, values = (var_5219_cast, var_5330_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, 64, 1, 256]> var_5369_cast = einsum(equation = var_5369_equation_0, values = (var_5223_cast, var_5331_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, 64, 1, 256]> var_5371_cast = einsum(equation = var_5371_equation_0, values = (var_5227_cast, var_5332_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, 64, 1, 256]> var_5373_cast = einsum(equation = var_5373_equation_0, values = (var_5231_cast, var_5333_cast))[name = tensor<string, []>("op_5373_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_3883, interleave = input_159_interleave_0, values = (var_5335_cast, var_5337_cast, var_5339_cast, var_5341_cast, var_5343_cast, var_5345_cast, var_5347_cast, var_5349_cast, var_5351_cast, var_5353_cast, var_5355_cast, var_5357_cast, var_5359_cast, var_5361_cast, var_5363_cast, var_5365_cast, var_5367_cast, var_5369_cast, var_5371_cast, var_5373_cast))[name = tensor<string, []>("input_159_cast")];
            tensor<int32, [2]> var_5379 = const()[name = tensor<string, []>("op_5379"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5381 = const()[name = tensor<string, []>("op_5381"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5383_pad_type_0 = const()[name = tensor<string, []>("op_5383_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5383_pad_0 = const()[name = tensor<string, []>("op_5383_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37191104))), lut = tensor<fp16, [4]>([-0x1.4e8p-4, -0x1.90cp-6, 0x1.8ep-6, 0x1.4d4p-4]), 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, []>(37600768)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5383_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_5381, groups = var_3883, pad = var_5383_pad_0, pad_type = var_5383_pad_type_0, strides = var_5379, 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_5383_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_33_cast = add(x = var_5383_cast, y = inputs_31_cast)[name = tensor<string, []>("inputs_33_cast")];
            tensor<int32, [1]> var_5387 = const()[name = tensor<string, []>("op_5387"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_33_cast = reduce_mean(axes = var_5387, keep_dims = var_3878, 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_5391 = const()[name = tensor<string, []>("op_5391"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5392_cast = reduce_mean(axes = var_5391, keep_dims = var_3878, x = zero_mean_sq_33_cast)[name = tensor<string, []>("op_5392_cast")];
            tensor<fp16, []> var_5393_to_fp16 = const()[name = tensor<string, []>("op_5393_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5394_cast = add(x = var_5392_cast, y = var_5393_to_fp16)[name = tensor<string, []>("op_5394_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_5394_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_5398_to_fp16 = const()[name = tensor<string, []>("op_5398_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37603392)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5399_cast = add(x = out_33_cast, y = var_5398_to_fp16)[name = tensor<string, []>("op_5399_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5401_to_fp16 = const()[name = tensor<string, []>("op_5401_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37606016)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_101_cast = mul(x = var_5399_cast, y = var_5401_to_fp16)[name = tensor<string, []>("hidden_states_101_cast")];
            tensor<int32, [2]> var_5408 = const()[name = tensor<string, []>("op_5408"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5410 = const()[name = tensor<string, []>("op_5410"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37608640))), lut = tensor<fp16, [4]>([-0x1.2fp-4, -0x1.69cp-6, 0x1.69cp-6, 0x1.2f4p-4]), 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_5410, groups = var_3883, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = var_5408, 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_5414 = const()[name = tensor<string, []>("op_5414"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5416 = const()[name = tensor<string, []>("op_5416"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_45_pad_type_0 = const()[name = tensor<string, []>("k_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_45_pad_0 = const()[name = tensor<string, []>("k_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38018304))), lut = tensor<fp16, [4]>([-0x1.3cp-4, -0x1.76p-6, 0x1.75cp-6, 0x1.3cp-4]), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_45_cast = conv(dilations = var_5416, groups = var_3883, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = var_5414, 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_5420 = const()[name = tensor<string, []>("op_5420"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5422 = const()[name = tensor<string, []>("op_5422"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_23_pad_type_0 = const()[name = tensor<string, []>("v_23_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_23_pad_0 = const()[name = tensor<string, []>("v_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38346048))), lut = tensor<fp16, [4]>([-0x1.bfcp-5, -0x1.06p-6, 0x1.088p-6, 0x1.c2p-5]), name = tensor<string, []>("down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_23_cast = conv(dilations = var_5422, groups = var_3883, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = var_5420, 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_5426_begin_0 = const()[name = tensor<string, []>("op_5426_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5426_end_0 = const()[name = tensor<string, []>("op_5426_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_5426_end_mask_0 = const()[name = tensor<string, []>("op_5426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5426_cast = slice_by_index(begin = var_5426_begin_0, end = var_5426_end_0, end_mask = var_5426_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5426_cast")];
            tensor<int32, [4]> var_5430_begin_0 = const()[name = tensor<string, []>("op_5430_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5430_end_0 = const()[name = tensor<string, []>("op_5430_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_5430_end_mask_0 = const()[name = tensor<string, []>("op_5430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5430_cast = slice_by_index(begin = var_5430_begin_0, end = var_5430_end_0, end_mask = var_5430_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5430_cast")];
            tensor<int32, [4]> var_5434_begin_0 = const()[name = tensor<string, []>("op_5434_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5434_end_0 = const()[name = tensor<string, []>("op_5434_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_5434_end_mask_0 = const()[name = tensor<string, []>("op_5434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5434_cast = slice_by_index(begin = var_5434_begin_0, end = var_5434_end_0, end_mask = var_5434_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5434_cast")];
            tensor<int32, [4]> var_5438_begin_0 = const()[name = tensor<string, []>("op_5438_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5438_end_0 = const()[name = tensor<string, []>("op_5438_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_5438_end_mask_0 = const()[name = tensor<string, []>("op_5438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5438_cast = slice_by_index(begin = var_5438_begin_0, end = var_5438_end_0, end_mask = var_5438_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5438_cast")];
            tensor<int32, [4]> var_5442_begin_0 = const()[name = tensor<string, []>("op_5442_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5442_end_0 = const()[name = tensor<string, []>("op_5442_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_5442_end_mask_0 = const()[name = tensor<string, []>("op_5442_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5442_cast = slice_by_index(begin = var_5442_begin_0, end = var_5442_end_0, end_mask = var_5442_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5442_cast")];
            tensor<int32, [4]> var_5446_begin_0 = const()[name = tensor<string, []>("op_5446_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5446_end_0 = const()[name = tensor<string, []>("op_5446_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_5446_end_mask_0 = const()[name = tensor<string, []>("op_5446_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5446_cast = slice_by_index(begin = var_5446_begin_0, end = var_5446_end_0, end_mask = var_5446_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5446_cast")];
            tensor<int32, [4]> var_5450_begin_0 = const()[name = tensor<string, []>("op_5450_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5450_end_0 = const()[name = tensor<string, []>("op_5450_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_5450_end_mask_0 = const()[name = tensor<string, []>("op_5450_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5450_cast = slice_by_index(begin = var_5450_begin_0, end = var_5450_end_0, end_mask = var_5450_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5450_cast")];
            tensor<int32, [4]> var_5454_begin_0 = const()[name = tensor<string, []>("op_5454_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5454_end_0 = const()[name = tensor<string, []>("op_5454_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_5454_end_mask_0 = const()[name = tensor<string, []>("op_5454_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5454_cast = slice_by_index(begin = var_5454_begin_0, end = var_5454_end_0, end_mask = var_5454_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5454_cast")];
            tensor<int32, [4]> var_5458_begin_0 = const()[name = tensor<string, []>("op_5458_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5458_end_0 = const()[name = tensor<string, []>("op_5458_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_5458_end_mask_0 = const()[name = tensor<string, []>("op_5458_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5458_cast = slice_by_index(begin = var_5458_begin_0, end = var_5458_end_0, end_mask = var_5458_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5458_cast")];
            tensor<int32, [4]> var_5462_begin_0 = const()[name = tensor<string, []>("op_5462_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5462_end_0 = const()[name = tensor<string, []>("op_5462_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_5462_end_mask_0 = const()[name = tensor<string, []>("op_5462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5462_cast = slice_by_index(begin = var_5462_begin_0, end = var_5462_end_0, end_mask = var_5462_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5462_cast")];
            tensor<int32, [4]> var_5466_begin_0 = const()[name = tensor<string, []>("op_5466_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5466_end_0 = const()[name = tensor<string, []>("op_5466_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_5466_end_mask_0 = const()[name = tensor<string, []>("op_5466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5466_cast = slice_by_index(begin = var_5466_begin_0, end = var_5466_end_0, end_mask = var_5466_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5466_cast")];
            tensor<int32, [4]> var_5470_begin_0 = const()[name = tensor<string, []>("op_5470_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5470_end_0 = const()[name = tensor<string, []>("op_5470_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_5470_end_mask_0 = const()[name = tensor<string, []>("op_5470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5470_cast = slice_by_index(begin = var_5470_begin_0, end = var_5470_end_0, end_mask = var_5470_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5470_cast")];
            tensor<int32, [4]> var_5474_begin_0 = const()[name = tensor<string, []>("op_5474_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5474_end_0 = const()[name = tensor<string, []>("op_5474_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_5474_end_mask_0 = const()[name = tensor<string, []>("op_5474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5474_cast = slice_by_index(begin = var_5474_begin_0, end = var_5474_end_0, end_mask = var_5474_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5474_cast")];
            tensor<int32, [4]> var_5478_begin_0 = const()[name = tensor<string, []>("op_5478_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5478_end_0 = const()[name = tensor<string, []>("op_5478_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_5478_end_mask_0 = const()[name = tensor<string, []>("op_5478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5478_cast = slice_by_index(begin = var_5478_begin_0, end = var_5478_end_0, end_mask = var_5478_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5478_cast")];
            tensor<int32, [4]> var_5482_begin_0 = const()[name = tensor<string, []>("op_5482_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5482_end_0 = const()[name = tensor<string, []>("op_5482_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_5482_end_mask_0 = const()[name = tensor<string, []>("op_5482_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5482_cast = slice_by_index(begin = var_5482_begin_0, end = var_5482_end_0, end_mask = var_5482_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5482_cast")];
            tensor<int32, [4]> var_5486_begin_0 = const()[name = tensor<string, []>("op_5486_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5486_end_0 = const()[name = tensor<string, []>("op_5486_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_5486_end_mask_0 = const()[name = tensor<string, []>("op_5486_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5486_cast = slice_by_index(begin = var_5486_begin_0, end = var_5486_end_0, end_mask = var_5486_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5486_cast")];
            tensor<int32, [4]> var_5490_begin_0 = const()[name = tensor<string, []>("op_5490_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5490_end_0 = const()[name = tensor<string, []>("op_5490_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_5490_end_mask_0 = const()[name = tensor<string, []>("op_5490_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5490_cast = slice_by_index(begin = var_5490_begin_0, end = var_5490_end_0, end_mask = var_5490_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5490_cast")];
            tensor<int32, [4]> var_5494_begin_0 = const()[name = tensor<string, []>("op_5494_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5494_end_0 = const()[name = tensor<string, []>("op_5494_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_5494_end_mask_0 = const()[name = tensor<string, []>("op_5494_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5494_cast = slice_by_index(begin = var_5494_begin_0, end = var_5494_end_0, end_mask = var_5494_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5494_cast")];
            tensor<int32, [4]> var_5498_begin_0 = const()[name = tensor<string, []>("op_5498_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5498_end_0 = const()[name = tensor<string, []>("op_5498_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_5498_end_mask_0 = const()[name = tensor<string, []>("op_5498_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5498_cast = slice_by_index(begin = var_5498_begin_0, end = var_5498_end_0, end_mask = var_5498_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5498_cast")];
            tensor<int32, [4]> var_5502_begin_0 = const()[name = tensor<string, []>("op_5502_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5502_end_0 = const()[name = tensor<string, []>("op_5502_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_5502_end_mask_0 = const()[name = tensor<string, []>("op_5502_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_5502_cast = slice_by_index(begin = var_5502_begin_0, end = var_5502_end_0, end_mask = var_5502_end_mask_0, x = q_23_cast)[name = tensor<string, []>("op_5502_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_5509_begin_0 = const()[name = tensor<string, []>("op_5509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5509_end_0 = const()[name = tensor<string, []>("op_5509_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_5509_end_mask_0 = const()[name = tensor<string, []>("op_5509_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, 64]> var_5509_cast = slice_by_index(begin = var_5509_begin_0, end = var_5509_end_0, end_mask = var_5509_end_mask_0, x = transpose_20)[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, 0, 0, 64])];
            tensor<int32, [4]> var_5513_end_0 = const()[name = tensor<string, []>("op_5513_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_5513_end_mask_0 = const()[name = tensor<string, []>("op_5513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5513_cast = slice_by_index(begin = var_5513_begin_0, end = var_5513_end_0, end_mask = var_5513_end_mask_0, x = transpose_20)[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, 0, 0, 128])];
            tensor<int32, [4]> var_5517_end_0 = const()[name = tensor<string, []>("op_5517_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_5517_end_mask_0 = const()[name = tensor<string, []>("op_5517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5517_cast = slice_by_index(begin = var_5517_begin_0, end = var_5517_end_0, end_mask = var_5517_end_mask_0, x = transpose_20)[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, 0, 0, 192])];
            tensor<int32, [4]> var_5521_end_0 = const()[name = tensor<string, []>("op_5521_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_5521_end_mask_0 = const()[name = tensor<string, []>("op_5521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5521_cast = slice_by_index(begin = var_5521_begin_0, end = var_5521_end_0, end_mask = var_5521_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5521_cast")];
            tensor<int32, [4]> var_5525_begin_0 = const()[name = tensor<string, []>("op_5525_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_5525_end_0 = const()[name = tensor<string, []>("op_5525_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_5525_end_mask_0 = const()[name = tensor<string, []>("op_5525_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5525_cast = slice_by_index(begin = var_5525_begin_0, end = var_5525_end_0, end_mask = var_5525_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5525_cast")];
            tensor<int32, [4]> var_5529_begin_0 = const()[name = tensor<string, []>("op_5529_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_5529_end_0 = const()[name = tensor<string, []>("op_5529_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_5529_end_mask_0 = const()[name = tensor<string, []>("op_5529_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5529_cast = slice_by_index(begin = var_5529_begin_0, end = var_5529_end_0, end_mask = var_5529_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5529_cast")];
            tensor<int32, [4]> var_5533_begin_0 = const()[name = tensor<string, []>("op_5533_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_5533_end_0 = const()[name = tensor<string, []>("op_5533_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_5533_end_mask_0 = const()[name = tensor<string, []>("op_5533_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5533_cast = slice_by_index(begin = var_5533_begin_0, end = var_5533_end_0, end_mask = var_5533_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5533_cast")];
            tensor<int32, [4]> var_5537_begin_0 = const()[name = tensor<string, []>("op_5537_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_5537_end_0 = const()[name = tensor<string, []>("op_5537_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_5537_end_mask_0 = const()[name = tensor<string, []>("op_5537_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5537_cast = slice_by_index(begin = var_5537_begin_0, end = var_5537_end_0, end_mask = var_5537_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5537_cast")];
            tensor<int32, [4]> var_5541_begin_0 = const()[name = tensor<string, []>("op_5541_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_5541_end_0 = const()[name = tensor<string, []>("op_5541_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_5541_end_mask_0 = const()[name = tensor<string, []>("op_5541_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5541_cast = slice_by_index(begin = var_5541_begin_0, end = var_5541_end_0, end_mask = var_5541_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5541_cast")];
            tensor<int32, [4]> var_5545_begin_0 = const()[name = tensor<string, []>("op_5545_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_5545_end_0 = const()[name = tensor<string, []>("op_5545_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_5545_end_mask_0 = const()[name = tensor<string, []>("op_5545_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5545_cast = slice_by_index(begin = var_5545_begin_0, end = var_5545_end_0, end_mask = var_5545_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5545_cast")];
            tensor<int32, [4]> var_5549_begin_0 = const()[name = tensor<string, []>("op_5549_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_5549_end_0 = const()[name = tensor<string, []>("op_5549_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_5549_end_mask_0 = const()[name = tensor<string, []>("op_5549_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5549_cast = slice_by_index(begin = var_5549_begin_0, end = var_5549_end_0, end_mask = var_5549_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5549_cast")];
            tensor<int32, [4]> var_5553_begin_0 = const()[name = tensor<string, []>("op_5553_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_5553_end_0 = const()[name = tensor<string, []>("op_5553_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_5553_end_mask_0 = const()[name = tensor<string, []>("op_5553_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5553_cast = slice_by_index(begin = var_5553_begin_0, end = var_5553_end_0, end_mask = var_5553_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5553_cast")];
            tensor<int32, [4]> var_5557_begin_0 = const()[name = tensor<string, []>("op_5557_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_5557_end_0 = const()[name = tensor<string, []>("op_5557_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_5557_end_mask_0 = const()[name = tensor<string, []>("op_5557_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5557_cast = slice_by_index(begin = var_5557_begin_0, end = var_5557_end_0, end_mask = var_5557_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5557_cast")];
            tensor<int32, [4]> var_5561_begin_0 = const()[name = tensor<string, []>("op_5561_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_5561_end_0 = const()[name = tensor<string, []>("op_5561_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_5561_end_mask_0 = const()[name = tensor<string, []>("op_5561_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5561_cast = slice_by_index(begin = var_5561_begin_0, end = var_5561_end_0, end_mask = var_5561_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5561_cast")];
            tensor<int32, [4]> var_5565_begin_0 = const()[name = tensor<string, []>("op_5565_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_5565_end_0 = const()[name = tensor<string, []>("op_5565_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_5565_end_mask_0 = const()[name = tensor<string, []>("op_5565_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5565_cast = slice_by_index(begin = var_5565_begin_0, end = var_5565_end_0, end_mask = var_5565_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5565_cast")];
            tensor<int32, [4]> var_5569_begin_0 = const()[name = tensor<string, []>("op_5569_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_5569_end_0 = const()[name = tensor<string, []>("op_5569_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_5569_end_mask_0 = const()[name = tensor<string, []>("op_5569_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5569_cast = slice_by_index(begin = var_5569_begin_0, end = var_5569_end_0, end_mask = var_5569_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5569_cast")];
            tensor<int32, [4]> var_5573_begin_0 = const()[name = tensor<string, []>("op_5573_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_5573_end_0 = const()[name = tensor<string, []>("op_5573_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_5573_end_mask_0 = const()[name = tensor<string, []>("op_5573_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5573_cast = slice_by_index(begin = var_5573_begin_0, end = var_5573_end_0, end_mask = var_5573_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5573_cast")];
            tensor<int32, [4]> var_5577_begin_0 = const()[name = tensor<string, []>("op_5577_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_5577_end_0 = const()[name = tensor<string, []>("op_5577_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_5577_end_mask_0 = const()[name = tensor<string, []>("op_5577_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5577_cast = slice_by_index(begin = var_5577_begin_0, end = var_5577_end_0, end_mask = var_5577_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5577_cast")];
            tensor<int32, [4]> var_5581_begin_0 = const()[name = tensor<string, []>("op_5581_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_5581_end_0 = const()[name = tensor<string, []>("op_5581_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_5581_end_mask_0 = const()[name = tensor<string, []>("op_5581_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5581_cast = slice_by_index(begin = var_5581_begin_0, end = var_5581_end_0, end_mask = var_5581_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5581_cast")];
            tensor<int32, [4]> var_5585_begin_0 = const()[name = tensor<string, []>("op_5585_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_5585_end_0 = const()[name = tensor<string, []>("op_5585_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_5585_end_mask_0 = const()[name = tensor<string, []>("op_5585_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_5585_cast = slice_by_index(begin = var_5585_begin_0, end = var_5585_end_0, end_mask = var_5585_end_mask_0, x = transpose_20)[name = tensor<string, []>("op_5585_cast")];
            tensor<int32, [4]> var_5587_begin_0 = const()[name = tensor<string, []>("op_5587_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_5587_end_0 = const()[name = tensor<string, []>("op_5587_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_5587_end_mask_0 = const()[name = tensor<string, []>("op_5587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5587_cast = slice_by_index(begin = var_5587_begin_0, end = var_5587_end_0, end_mask = var_5587_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5587_cast")];
            tensor<int32, [4]> var_5591_begin_0 = const()[name = tensor<string, []>("op_5591_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_5591_end_0 = const()[name = tensor<string, []>("op_5591_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_5591_end_mask_0 = const()[name = tensor<string, []>("op_5591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5591_cast = slice_by_index(begin = var_5591_begin_0, end = var_5591_end_0, end_mask = var_5591_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5591_cast")];
            tensor<int32, [4]> var_5595_begin_0 = const()[name = tensor<string, []>("op_5595_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_5595_end_0 = const()[name = tensor<string, []>("op_5595_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_5595_end_mask_0 = const()[name = tensor<string, []>("op_5595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5595_cast = slice_by_index(begin = var_5595_begin_0, end = var_5595_end_0, end_mask = var_5595_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5595_cast")];
            tensor<int32, [4]> var_5599_begin_0 = const()[name = tensor<string, []>("op_5599_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_5599_end_0 = const()[name = tensor<string, []>("op_5599_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_5599_end_mask_0 = const()[name = tensor<string, []>("op_5599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5599_cast = slice_by_index(begin = var_5599_begin_0, end = var_5599_end_0, end_mask = var_5599_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5599_cast")];
            tensor<int32, [4]> var_5603_begin_0 = const()[name = tensor<string, []>("op_5603_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_5603_end_0 = const()[name = tensor<string, []>("op_5603_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_5603_end_mask_0 = const()[name = tensor<string, []>("op_5603_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5603_cast = slice_by_index(begin = var_5603_begin_0, end = var_5603_end_0, end_mask = var_5603_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5603_cast")];
            tensor<int32, [4]> var_5607_begin_0 = const()[name = tensor<string, []>("op_5607_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_5607_end_0 = const()[name = tensor<string, []>("op_5607_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_5607_end_mask_0 = const()[name = tensor<string, []>("op_5607_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5607_cast = slice_by_index(begin = var_5607_begin_0, end = var_5607_end_0, end_mask = var_5607_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5607_cast")];
            tensor<int32, [4]> var_5611_begin_0 = const()[name = tensor<string, []>("op_5611_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_5611_end_0 = const()[name = tensor<string, []>("op_5611_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_5611_end_mask_0 = const()[name = tensor<string, []>("op_5611_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5611_cast = slice_by_index(begin = var_5611_begin_0, end = var_5611_end_0, end_mask = var_5611_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5611_cast")];
            tensor<int32, [4]> var_5615_begin_0 = const()[name = tensor<string, []>("op_5615_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_5615_end_0 = const()[name = tensor<string, []>("op_5615_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_5615_end_mask_0 = const()[name = tensor<string, []>("op_5615_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5615_cast = slice_by_index(begin = var_5615_begin_0, end = var_5615_end_0, end_mask = var_5615_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5615_cast")];
            tensor<int32, [4]> var_5619_begin_0 = const()[name = tensor<string, []>("op_5619_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_5619_end_0 = const()[name = tensor<string, []>("op_5619_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_5619_end_mask_0 = const()[name = tensor<string, []>("op_5619_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5619_cast = slice_by_index(begin = var_5619_begin_0, end = var_5619_end_0, end_mask = var_5619_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5619_cast")];
            tensor<int32, [4]> var_5623_begin_0 = const()[name = tensor<string, []>("op_5623_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_5623_end_0 = const()[name = tensor<string, []>("op_5623_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_5623_end_mask_0 = const()[name = tensor<string, []>("op_5623_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5623_cast = slice_by_index(begin = var_5623_begin_0, end = var_5623_end_0, end_mask = var_5623_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5623_cast")];
            tensor<int32, [4]> var_5627_begin_0 = const()[name = tensor<string, []>("op_5627_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_5627_end_0 = const()[name = tensor<string, []>("op_5627_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_5627_end_mask_0 = const()[name = tensor<string, []>("op_5627_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5627_cast = slice_by_index(begin = var_5627_begin_0, end = var_5627_end_0, end_mask = var_5627_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5627_cast")];
            tensor<int32, [4]> var_5631_begin_0 = const()[name = tensor<string, []>("op_5631_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_5631_end_0 = const()[name = tensor<string, []>("op_5631_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_5631_end_mask_0 = const()[name = tensor<string, []>("op_5631_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5631_cast = slice_by_index(begin = var_5631_begin_0, end = var_5631_end_0, end_mask = var_5631_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5631_cast")];
            tensor<int32, [4]> var_5635_begin_0 = const()[name = tensor<string, []>("op_5635_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_5635_end_0 = const()[name = tensor<string, []>("op_5635_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_5635_end_mask_0 = const()[name = tensor<string, []>("op_5635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5635_cast = slice_by_index(begin = var_5635_begin_0, end = var_5635_end_0, end_mask = var_5635_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5635_cast")];
            tensor<int32, [4]> var_5639_begin_0 = const()[name = tensor<string, []>("op_5639_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_5639_end_0 = const()[name = tensor<string, []>("op_5639_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_5639_end_mask_0 = const()[name = tensor<string, []>("op_5639_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5639_cast = slice_by_index(begin = var_5639_begin_0, end = var_5639_end_0, end_mask = var_5639_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5639_cast")];
            tensor<int32, [4]> var_5643_begin_0 = const()[name = tensor<string, []>("op_5643_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_5643_end_0 = const()[name = tensor<string, []>("op_5643_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_5643_end_mask_0 = const()[name = tensor<string, []>("op_5643_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5643_cast = slice_by_index(begin = var_5643_begin_0, end = var_5643_end_0, end_mask = var_5643_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5643_cast")];
            tensor<int32, [4]> var_5647_begin_0 = const()[name = tensor<string, []>("op_5647_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_5647_end_0 = const()[name = tensor<string, []>("op_5647_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_5647_end_mask_0 = const()[name = tensor<string, []>("op_5647_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5647_cast = slice_by_index(begin = var_5647_begin_0, end = var_5647_end_0, end_mask = var_5647_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5647_cast")];
            tensor<int32, [4]> var_5651_begin_0 = const()[name = tensor<string, []>("op_5651_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_5651_end_0 = const()[name = tensor<string, []>("op_5651_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_5651_end_mask_0 = const()[name = tensor<string, []>("op_5651_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5651_cast = slice_by_index(begin = var_5651_begin_0, end = var_5651_end_0, end_mask = var_5651_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5651_cast")];
            tensor<int32, [4]> var_5655_begin_0 = const()[name = tensor<string, []>("op_5655_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_5655_end_0 = const()[name = tensor<string, []>("op_5655_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_5655_end_mask_0 = const()[name = tensor<string, []>("op_5655_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5655_cast = slice_by_index(begin = var_5655_begin_0, end = var_5655_end_0, end_mask = var_5655_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5655_cast")];
            tensor<int32, [4]> var_5659_begin_0 = const()[name = tensor<string, []>("op_5659_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_5659_end_0 = const()[name = tensor<string, []>("op_5659_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_5659_end_mask_0 = const()[name = tensor<string, []>("op_5659_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5659_cast = slice_by_index(begin = var_5659_begin_0, end = var_5659_end_0, end_mask = var_5659_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5659_cast")];
            tensor<int32, [4]> var_5663_begin_0 = const()[name = tensor<string, []>("op_5663_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_5663_end_0 = const()[name = tensor<string, []>("op_5663_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_5663_end_mask_0 = const()[name = tensor<string, []>("op_5663_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_5663_cast = slice_by_index(begin = var_5663_begin_0, end = var_5663_end_0, end_mask = var_5663_end_mask_0, x = v_23_cast)[name = tensor<string, []>("op_5663_cast")];
            tensor<string, []> var_5667_equation_0 = const()[name = tensor<string, []>("op_5667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5667_cast = einsum(equation = var_5667_equation_0, values = (var_5509_cast, var_5426_cast))[name = tensor<string, []>("op_5667_cast")];
            tensor<fp16, []> var_5668_to_fp16 = const()[name = tensor<string, []>("op_5668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_121_cast = mul(x = var_5667_cast, y = var_5668_to_fp16)[name = tensor<string, []>("aw_121_cast")];
            tensor<string, []> var_5671_equation_0 = const()[name = tensor<string, []>("op_5671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5671_cast = einsum(equation = var_5671_equation_0, values = (var_5513_cast, var_5430_cast))[name = tensor<string, []>("op_5671_cast")];
            tensor<fp16, []> var_5672_to_fp16 = const()[name = tensor<string, []>("op_5672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_123_cast = mul(x = var_5671_cast, y = var_5672_to_fp16)[name = tensor<string, []>("aw_123_cast")];
            tensor<string, []> var_5675_equation_0 = const()[name = tensor<string, []>("op_5675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5675_cast = einsum(equation = var_5675_equation_0, values = (var_5517_cast, var_5434_cast))[name = tensor<string, []>("op_5675_cast")];
            tensor<fp16, []> var_5676_to_fp16 = const()[name = tensor<string, []>("op_5676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_125_cast = mul(x = var_5675_cast, y = var_5676_to_fp16)[name = tensor<string, []>("aw_125_cast")];
            tensor<string, []> var_5679_equation_0 = const()[name = tensor<string, []>("op_5679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5679_cast = einsum(equation = var_5679_equation_0, values = (var_5521_cast, var_5438_cast))[name = tensor<string, []>("op_5679_cast")];
            tensor<fp16, []> var_5680_to_fp16 = const()[name = tensor<string, []>("op_5680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_127_cast = mul(x = var_5679_cast, y = var_5680_to_fp16)[name = tensor<string, []>("aw_127_cast")];
            tensor<string, []> var_5683_equation_0 = const()[name = tensor<string, []>("op_5683_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5683_cast = einsum(equation = var_5683_equation_0, values = (var_5525_cast, var_5442_cast))[name = tensor<string, []>("op_5683_cast")];
            tensor<fp16, []> var_5684_to_fp16 = const()[name = tensor<string, []>("op_5684_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_129_cast = mul(x = var_5683_cast, y = var_5684_to_fp16)[name = tensor<string, []>("aw_129_cast")];
            tensor<string, []> var_5687_equation_0 = const()[name = tensor<string, []>("op_5687_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5687_cast = einsum(equation = var_5687_equation_0, values = (var_5529_cast, var_5446_cast))[name = tensor<string, []>("op_5687_cast")];
            tensor<fp16, []> var_5688_to_fp16 = const()[name = tensor<string, []>("op_5688_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_131_cast = mul(x = var_5687_cast, y = var_5688_to_fp16)[name = tensor<string, []>("aw_131_cast")];
            tensor<string, []> var_5691_equation_0 = const()[name = tensor<string, []>("op_5691_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5691_cast = einsum(equation = var_5691_equation_0, values = (var_5533_cast, var_5450_cast))[name = tensor<string, []>("op_5691_cast")];
            tensor<fp16, []> var_5692_to_fp16 = const()[name = tensor<string, []>("op_5692_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_133_cast = mul(x = var_5691_cast, y = var_5692_to_fp16)[name = tensor<string, []>("aw_133_cast")];
            tensor<string, []> var_5695_equation_0 = const()[name = tensor<string, []>("op_5695_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5695_cast = einsum(equation = var_5695_equation_0, values = (var_5537_cast, var_5454_cast))[name = tensor<string, []>("op_5695_cast")];
            tensor<fp16, []> var_5696_to_fp16 = const()[name = tensor<string, []>("op_5696_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_135_cast = mul(x = var_5695_cast, y = var_5696_to_fp16)[name = tensor<string, []>("aw_135_cast")];
            tensor<string, []> var_5699_equation_0 = const()[name = tensor<string, []>("op_5699_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5699_cast = einsum(equation = var_5699_equation_0, values = (var_5541_cast, var_5458_cast))[name = tensor<string, []>("op_5699_cast")];
            tensor<fp16, []> var_5700_to_fp16 = const()[name = tensor<string, []>("op_5700_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_137_cast = mul(x = var_5699_cast, y = var_5700_to_fp16)[name = tensor<string, []>("aw_137_cast")];
            tensor<string, []> var_5703_equation_0 = const()[name = tensor<string, []>("op_5703_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5703_cast = einsum(equation = var_5703_equation_0, values = (var_5545_cast, var_5462_cast))[name = tensor<string, []>("op_5703_cast")];
            tensor<fp16, []> var_5704_to_fp16 = const()[name = tensor<string, []>("op_5704_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_139_cast = mul(x = var_5703_cast, y = var_5704_to_fp16)[name = tensor<string, []>("aw_139_cast")];
            tensor<string, []> var_5707_equation_0 = const()[name = tensor<string, []>("op_5707_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5707_cast = einsum(equation = var_5707_equation_0, values = (var_5549_cast, var_5466_cast))[name = tensor<string, []>("op_5707_cast")];
            tensor<fp16, []> var_5708_to_fp16 = const()[name = tensor<string, []>("op_5708_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_141_cast = mul(x = var_5707_cast, y = var_5708_to_fp16)[name = tensor<string, []>("aw_141_cast")];
            tensor<string, []> var_5711_equation_0 = const()[name = tensor<string, []>("op_5711_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5711_cast = einsum(equation = var_5711_equation_0, values = (var_5553_cast, var_5470_cast))[name = tensor<string, []>("op_5711_cast")];
            tensor<fp16, []> var_5712_to_fp16 = const()[name = tensor<string, []>("op_5712_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_143_cast = mul(x = var_5711_cast, y = var_5712_to_fp16)[name = tensor<string, []>("aw_143_cast")];
            tensor<string, []> var_5715_equation_0 = const()[name = tensor<string, []>("op_5715_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5715_cast = einsum(equation = var_5715_equation_0, values = (var_5557_cast, var_5474_cast))[name = tensor<string, []>("op_5715_cast")];
            tensor<fp16, []> var_5716_to_fp16 = const()[name = tensor<string, []>("op_5716_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_145_cast = mul(x = var_5715_cast, y = var_5716_to_fp16)[name = tensor<string, []>("aw_145_cast")];
            tensor<string, []> var_5719_equation_0 = const()[name = tensor<string, []>("op_5719_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5719_cast = einsum(equation = var_5719_equation_0, values = (var_5561_cast, var_5478_cast))[name = tensor<string, []>("op_5719_cast")];
            tensor<fp16, []> var_5720_to_fp16 = const()[name = tensor<string, []>("op_5720_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_147_cast = mul(x = var_5719_cast, y = var_5720_to_fp16)[name = tensor<string, []>("aw_147_cast")];
            tensor<string, []> var_5723_equation_0 = const()[name = tensor<string, []>("op_5723_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5723_cast = einsum(equation = var_5723_equation_0, values = (var_5565_cast, var_5482_cast))[name = tensor<string, []>("op_5723_cast")];
            tensor<fp16, []> var_5724_to_fp16 = const()[name = tensor<string, []>("op_5724_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_149_cast = mul(x = var_5723_cast, y = var_5724_to_fp16)[name = tensor<string, []>("aw_149_cast")];
            tensor<string, []> var_5727_equation_0 = const()[name = tensor<string, []>("op_5727_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5727_cast = einsum(equation = var_5727_equation_0, values = (var_5569_cast, var_5486_cast))[name = tensor<string, []>("op_5727_cast")];
            tensor<fp16, []> var_5728_to_fp16 = const()[name = tensor<string, []>("op_5728_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_151_cast = mul(x = var_5727_cast, y = var_5728_to_fp16)[name = tensor<string, []>("aw_151_cast")];
            tensor<string, []> var_5731_equation_0 = const()[name = tensor<string, []>("op_5731_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5731_cast = einsum(equation = var_5731_equation_0, values = (var_5573_cast, var_5490_cast))[name = tensor<string, []>("op_5731_cast")];
            tensor<fp16, []> var_5732_to_fp16 = const()[name = tensor<string, []>("op_5732_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_153_cast = mul(x = var_5731_cast, y = var_5732_to_fp16)[name = tensor<string, []>("aw_153_cast")];
            tensor<string, []> var_5735_equation_0 = const()[name = tensor<string, []>("op_5735_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5735_cast = einsum(equation = var_5735_equation_0, values = (var_5577_cast, var_5494_cast))[name = tensor<string, []>("op_5735_cast")];
            tensor<fp16, []> var_5736_to_fp16 = const()[name = tensor<string, []>("op_5736_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_155_cast = mul(x = var_5735_cast, y = var_5736_to_fp16)[name = tensor<string, []>("aw_155_cast")];
            tensor<string, []> var_5739_equation_0 = const()[name = tensor<string, []>("op_5739_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5739_cast = einsum(equation = var_5739_equation_0, values = (var_5581_cast, var_5498_cast))[name = tensor<string, []>("op_5739_cast")];
            tensor<fp16, []> var_5740_to_fp16 = const()[name = tensor<string, []>("op_5740_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_157_cast = mul(x = var_5739_cast, y = var_5740_to_fp16)[name = tensor<string, []>("aw_157_cast")];
            tensor<string, []> var_5743_equation_0 = const()[name = tensor<string, []>("op_5743_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_5743_cast = einsum(equation = var_5743_equation_0, values = (var_5585_cast, var_5502_cast))[name = tensor<string, []>("op_5743_cast")];
            tensor<fp16, []> var_5744_to_fp16 = const()[name = tensor<string, []>("op_5744_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_159_cast = mul(x = var_5743_cast, y = var_5744_to_fp16)[name = tensor<string, []>("aw_159_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5746_cast = softmax(axis = var_3883, x = aw_121_cast)[name = tensor<string, []>("op_5746_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5747_cast = softmax(axis = var_3883, x = aw_123_cast)[name = tensor<string, []>("op_5747_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5748_cast = softmax(axis = var_3883, x = aw_125_cast)[name = tensor<string, []>("op_5748_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5749_cast = softmax(axis = var_3883, x = aw_127_cast)[name = tensor<string, []>("op_5749_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5750_cast = softmax(axis = var_3883, x = aw_129_cast)[name = tensor<string, []>("op_5750_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5751_cast = softmax(axis = var_3883, x = aw_131_cast)[name = tensor<string, []>("op_5751_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5752_cast = softmax(axis = var_3883, x = aw_133_cast)[name = tensor<string, []>("op_5752_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5753_cast = softmax(axis = var_3883, x = aw_135_cast)[name = tensor<string, []>("op_5753_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5754_cast = softmax(axis = var_3883, x = aw_137_cast)[name = tensor<string, []>("op_5754_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5755_cast = softmax(axis = var_3883, x = aw_139_cast)[name = tensor<string, []>("op_5755_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5756_cast = softmax(axis = var_3883, x = aw_141_cast)[name = tensor<string, []>("op_5756_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5757_cast = softmax(axis = var_3883, x = aw_143_cast)[name = tensor<string, []>("op_5757_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5758_cast = softmax(axis = var_3883, x = aw_145_cast)[name = tensor<string, []>("op_5758_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5759_cast = softmax(axis = var_3883, x = aw_147_cast)[name = tensor<string, []>("op_5759_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5760_cast = softmax(axis = var_3883, x = aw_149_cast)[name = tensor<string, []>("op_5760_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5761_cast = softmax(axis = var_3883, x = aw_151_cast)[name = tensor<string, []>("op_5761_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5762_cast = softmax(axis = var_3883, x = aw_153_cast)[name = tensor<string, []>("op_5762_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5763_cast = softmax(axis = var_3883, x = aw_155_cast)[name = tensor<string, []>("op_5763_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5764_cast = softmax(axis = var_3883, x = aw_157_cast)[name = tensor<string, []>("op_5764_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_5765_cast = softmax(axis = var_3883, x = aw_159_cast)[name = tensor<string, []>("op_5765_cast")];
            tensor<string, []> var_5767_equation_0 = const()[name = tensor<string, []>("op_5767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5767_cast = einsum(equation = var_5767_equation_0, values = (var_5587_cast, var_5746_cast))[name = tensor<string, []>("op_5767_cast")];
            tensor<string, []> var_5769_equation_0 = const()[name = tensor<string, []>("op_5769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5769_cast = einsum(equation = var_5769_equation_0, values = (var_5591_cast, var_5747_cast))[name = tensor<string, []>("op_5769_cast")];
            tensor<string, []> var_5771_equation_0 = const()[name = tensor<string, []>("op_5771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5771_cast = einsum(equation = var_5771_equation_0, values = (var_5595_cast, var_5748_cast))[name = tensor<string, []>("op_5771_cast")];
            tensor<string, []> var_5773_equation_0 = const()[name = tensor<string, []>("op_5773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5773_cast = einsum(equation = var_5773_equation_0, values = (var_5599_cast, var_5749_cast))[name = tensor<string, []>("op_5773_cast")];
            tensor<string, []> var_5775_equation_0 = const()[name = tensor<string, []>("op_5775_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5775_cast = einsum(equation = var_5775_equation_0, values = (var_5603_cast, var_5750_cast))[name = tensor<string, []>("op_5775_cast")];
            tensor<string, []> var_5777_equation_0 = const()[name = tensor<string, []>("op_5777_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5777_cast = einsum(equation = var_5777_equation_0, values = (var_5607_cast, var_5751_cast))[name = tensor<string, []>("op_5777_cast")];
            tensor<string, []> var_5779_equation_0 = const()[name = tensor<string, []>("op_5779_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5779_cast = einsum(equation = var_5779_equation_0, values = (var_5611_cast, var_5752_cast))[name = tensor<string, []>("op_5779_cast")];
            tensor<string, []> var_5781_equation_0 = const()[name = tensor<string, []>("op_5781_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5781_cast = einsum(equation = var_5781_equation_0, values = (var_5615_cast, var_5753_cast))[name = tensor<string, []>("op_5781_cast")];
            tensor<string, []> var_5783_equation_0 = const()[name = tensor<string, []>("op_5783_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5783_cast = einsum(equation = var_5783_equation_0, values = (var_5619_cast, var_5754_cast))[name = tensor<string, []>("op_5783_cast")];
            tensor<string, []> var_5785_equation_0 = const()[name = tensor<string, []>("op_5785_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5785_cast = einsum(equation = var_5785_equation_0, values = (var_5623_cast, var_5755_cast))[name = tensor<string, []>("op_5785_cast")];
            tensor<string, []> var_5787_equation_0 = const()[name = tensor<string, []>("op_5787_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5787_cast = einsum(equation = var_5787_equation_0, values = (var_5627_cast, var_5756_cast))[name = tensor<string, []>("op_5787_cast")];
            tensor<string, []> var_5789_equation_0 = const()[name = tensor<string, []>("op_5789_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5789_cast = einsum(equation = var_5789_equation_0, values = (var_5631_cast, var_5757_cast))[name = tensor<string, []>("op_5789_cast")];
            tensor<string, []> var_5791_equation_0 = const()[name = tensor<string, []>("op_5791_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5791_cast = einsum(equation = var_5791_equation_0, values = (var_5635_cast, var_5758_cast))[name = tensor<string, []>("op_5791_cast")];
            tensor<string, []> var_5793_equation_0 = const()[name = tensor<string, []>("op_5793_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5793_cast = einsum(equation = var_5793_equation_0, values = (var_5639_cast, var_5759_cast))[name = tensor<string, []>("op_5793_cast")];
            tensor<string, []> var_5795_equation_0 = const()[name = tensor<string, []>("op_5795_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5795_cast = einsum(equation = var_5795_equation_0, values = (var_5643_cast, var_5760_cast))[name = tensor<string, []>("op_5795_cast")];
            tensor<string, []> var_5797_equation_0 = const()[name = tensor<string, []>("op_5797_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5797_cast = einsum(equation = var_5797_equation_0, values = (var_5647_cast, var_5761_cast))[name = tensor<string, []>("op_5797_cast")];
            tensor<string, []> var_5799_equation_0 = const()[name = tensor<string, []>("op_5799_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5799_cast = einsum(equation = var_5799_equation_0, values = (var_5651_cast, var_5762_cast))[name = tensor<string, []>("op_5799_cast")];
            tensor<string, []> var_5801_equation_0 = const()[name = tensor<string, []>("op_5801_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5801_cast = einsum(equation = var_5801_equation_0, values = (var_5655_cast, var_5763_cast))[name = tensor<string, []>("op_5801_cast")];
            tensor<string, []> var_5803_equation_0 = const()[name = tensor<string, []>("op_5803_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5803_cast = einsum(equation = var_5803_equation_0, values = (var_5659_cast, var_5764_cast))[name = tensor<string, []>("op_5803_cast")];
            tensor<string, []> var_5805_equation_0 = const()[name = tensor<string, []>("op_5805_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_5805_cast = einsum(equation = var_5805_equation_0, values = (var_5663_cast, var_5765_cast))[name = tensor<string, []>("op_5805_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_3883, interleave = input_161_interleave_0, values = (var_5767_cast, var_5769_cast, var_5771_cast, var_5773_cast, var_5775_cast, var_5777_cast, var_5779_cast, var_5781_cast, var_5783_cast, var_5785_cast, var_5787_cast, var_5789_cast, var_5791_cast, var_5793_cast, var_5795_cast, var_5797_cast, var_5799_cast, var_5801_cast, var_5803_cast, var_5805_cast))[name = tensor<string, []>("input_161_cast")];
            tensor<int32, [2]> var_5811 = const()[name = tensor<string, []>("op_5811"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5813 = const()[name = tensor<string, []>("op_5813"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5815_pad_type_0 = const()[name = tensor<string, []>("op_5815_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5815_pad_0 = const()[name = tensor<string, []>("op_5815_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38673792))), lut = tensor<fp16, [4]>([-0x1.4d4p-5, -0x1.8ap-7, 0x1.938p-7, 0x1.4fcp-5]), 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, []>(39083456)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5815_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_5813, groups = var_3883, pad = var_5815_pad_0, pad_type = var_5815_pad_type_0, strides = var_5811, 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_5815_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_35_cast = add(x = var_5815_cast, y = inputs_33_cast)[name = tensor<string, []>("inputs_35_cast")];
            tensor<int32, [1]> var_5819 = const()[name = tensor<string, []>("op_5819"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_35_cast = reduce_mean(axes = var_5819, keep_dims = var_3878, 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_5823 = const()[name = tensor<string, []>("op_5823"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_5824_cast = reduce_mean(axes = var_5823, keep_dims = var_3878, x = zero_mean_sq_35_cast)[name = tensor<string, []>("op_5824_cast")];
            tensor<fp16, []> var_5825_to_fp16 = const()[name = tensor<string, []>("op_5825_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_5826_cast = add(x = var_5824_cast, y = var_5825_to_fp16)[name = tensor<string, []>("op_5826_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_5826_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_5830_to_fp16 = const()[name = tensor<string, []>("op_5830_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39086080)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5831_cast = add(x = out_35_cast, y = var_5830_to_fp16)[name = tensor<string, []>("op_5831_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_5833_to_fp16 = const()[name = tensor<string, []>("op_5833_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39088704)))];
            tensor<fp16, [2, 1280, 1, 256]> input_163_cast = mul(x = var_5831_cast, y = var_5833_to_fp16)[name = tensor<string, []>("input_163_cast")];
            tensor<int32, [2]> var_5841 = const()[name = tensor<string, []>("op_5841"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5843 = const()[name = tensor<string, []>("op_5843"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5845_pad_type_0 = const()[name = tensor<string, []>("op_5845_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5845_pad_0 = const()[name = tensor<string, []>("op_5845_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, [3276800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39091328))), lut = tensor<fp16, [4]>([-0x1.5dcp-4, -0x1.9ccp-6, 0x1.9fcp-6, 0x1.5fp-4]), 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, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42368192))), lut = tensor<fp16, [4]>([-0x1.484p-3, -0x1.1fcp-4, -0x1.adcp-7, 0x1.22cp-5]), 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_5845_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_5843, groups = var_3883, pad = var_5845_pad_0, pad_type = var_5845_pad_type_0, strides = var_5841, 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_5845_cast")];
            tensor<int32, [2]> var_5846_split_sizes_0 = const()[name = tensor<string, []>("op_5846_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_5846_axis_0 = const()[name = tensor<string, []>("op_5846_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_5846_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_5846_cast_1 = split(axis = var_5846_axis_0, split_sizes = var_5846_split_sizes_0, x = var_5845_cast)[name = tensor<string, []>("op_5846_cast")];
            tensor<string, []> var_5848_mode_0 = const()[name = tensor<string, []>("op_5848_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_5848_cast = gelu(mode = var_5848_mode_0, x = var_5846_cast_1)[name = tensor<string, []>("op_5848_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_165_cast = mul(x = var_5846_cast_0, y = var_5848_cast)[name = tensor<string, []>("input_165_cast")];
            tensor<int32, [2]> var_5852 = const()[name = tensor<string, []>("op_5852"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5854 = const()[name = tensor<string, []>("op_5854"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_5856_pad_type_0 = const()[name = tensor<string, []>("op_5856_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_5856_pad_0 = const()[name = tensor<string, []>("op_5856_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, [1638400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42370816))), lut = tensor<fp16, [4]>([-0x1.4acp-4, -0x1.8b4p-6, 0x1.89cp-6, 0x1.4acp-4]), 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, []>(44009280)))];
            tensor<fp16, [2, 1280, 1, 256]> var_5856_cast = conv(bias = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_5854, groups = var_3883, pad = var_5856_pad_0, pad_type = var_5856_pad_type_0, strides = var_5852, weight = down_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_165_cast)[name = tensor<string, []>("op_5856_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_105_cast = add(x = var_5856_cast, y = inputs_35_cast)[name = tensor<string, []>("hidden_states_105_cast")];
            tensor<int32, [4]> var_5858 = const()[name = tensor<string, []>("op_5858"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_167_cast = reshape(shape = var_5858, x = hidden_states_105_cast)[name = tensor<string, []>("input_167_cast")];
            tensor<int32, [2]> var_5862 = const()[name = tensor<string, []>("op_5862"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5864 = const()[name = tensor<string, []>("op_5864"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44011904))), lut = tensor<fp16, [4]>([-0x1.17cp-4, -0x1.4c8p-6, 0x1.53cp-6, 0x1.198p-4]), 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, []>(44421568)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_107_cast = conv(bias = down_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_5864, groups = var_3883, pad = hidden_states_107_pad_0, pad_type = hidden_states_107_pad_type_0, strides = var_5862, 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_5871 = const()[name = tensor<string, []>("op_5871"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_5873 = const()[name = tensor<string, []>("op_5873"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44424192))), lut = tensor<fp16, [4]>([-0x1.b6p-6, -0x1.f2cp-8, 0x1.0ep-7, 0x1.c5cp-6]), 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, []>(48110656)))];
            tensor<fp16, [2, 1280, 8, 8]> input_171_cast = conv(bias = down_blocks_2_downsamplers_0_conv_bias_to_fp16, dilations = var_5873, groups = var_3883, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = var_5871, weight = down_blocks_2_downsamplers_0_conv_weight_to_fp16_palettized, x = input_169_cast)[name = tensor<string, []>("input_171_cast")];
            tensor<int32, []> var_5885 = const()[name = tensor<string, []>("op_5885"), 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, []>(48113280)))];
            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, []>(48115904)))];
            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_5901 = const()[name = tensor<string, []>("op_5901"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5903 = const()[name = tensor<string, []>("op_5903"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48118528))), lut = tensor<fp16, [4]>([-0x1.608p-5, -0x1.408p-7, 0x1.484p-7, 0x1.6e8p-5]), 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, []>(51804992)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_109_cast = conv(bias = down_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_5903, groups = var_5885, pad = hidden_states_109_pad_0, pad_type = hidden_states_109_pad_type_0, strides = var_5901, 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_5909 = const()[name = tensor<string, []>("op_5909"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5911 = const()[name = tensor<string, []>("op_5911"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51807616))), lut = tensor<fp16, [4]>([-0x1.144p-5, -0x1.74p-9, 0x1.87cp-9, 0x1.538p-5]), 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, []>(52217280)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_13_cast = conv(bias = down_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_5911, groups = var_5885, pad = temb_13_pad_0, pad_type = temb_13_pad_type_0, strides = var_5909, 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, []>(52219904)))];
            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, []>(52222528)))];
            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_5921 = const()[name = tensor<string, []>("op_5921"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5923 = const()[name = tensor<string, []>("op_5923"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52225152))), lut = tensor<fp16, [4]>([-0x1.8acp-5, -0x1.5fcp-7, 0x1.73p-7, 0x1.91p-5]), 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, []>(55911616)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_111_cast = conv(bias = down_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_5923, groups = var_5885, pad = hidden_states_111_pad_0, pad_type = hidden_states_111_pad_type_0, strides = var_5921, 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, []>(55914240)))];
            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, []>(55916864)))];
            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_5938 = const()[name = tensor<string, []>("op_5938"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5940 = const()[name = tensor<string, []>("op_5940"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55919488))), lut = tensor<fp16, [4]>([-0x1.39cp-5, -0x1.104p-7, 0x1.0bp-7, 0x1.3dcp-5]), 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, []>(59605952)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_113_cast = conv(bias = down_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_5940, groups = var_5885, pad = hidden_states_113_pad_0, pad_type = hidden_states_113_pad_type_0, strides = var_5938, 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_5946 = const()[name = tensor<string, []>("op_5946"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5948 = const()[name = tensor<string, []>("op_5948"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59608576))), lut = tensor<fp16, [4]>([-0x1.0e4p-8, -0x1.e6p-14, 0x1.ffp-9, 0x1.05cp-5]), 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, []>(60018240)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_15_cast = conv(bias = down_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_5948, groups = var_5885, pad = temb_15_pad_0, pad_type = temb_15_pad_type_0, strides = var_5946, 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, []>(60020864)))];
            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, []>(60023488)))];
            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_5958 = const()[name = tensor<string, []>("op_5958"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_5960 = const()[name = tensor<string, []>("op_5960"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60026112))), lut = tensor<fp16, [4]>([-0x1.4bp-5, -0x1.0b8p-7, 0x1.258p-7, 0x1.5ap-5]), 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, []>(63712576)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_115_cast = conv(bias = down_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_5960, groups = var_5885, pad = hidden_states_115_pad_0, pad_type = hidden_states_115_pad_type_0, strides = var_5958, 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_5995 = const()[name = tensor<string, []>("op_5995"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_6000 = const()[name = tensor<string, []>("op_6000"), 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, []>(63715200)))];
            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, []>(63717824)))];
            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_6018 = const()[name = tensor<string, []>("op_6018"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6020 = const()[name = tensor<string, []>("op_6020"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63720448))), lut = tensor<fp16, [4]>([-0x1.21cp-5, -0x1.adp-8, 0x1.abcp-8, 0x1.25p-5]), 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, []>(67406912)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_117_cast = conv(bias = mid_block_resnets_0_conv1_bias_to_fp16, dilations = var_6020, groups = var_6000, pad = hidden_states_117_pad_0, pad_type = hidden_states_117_pad_type_0, strides = var_6018, 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_6026 = const()[name = tensor<string, []>("op_6026"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6028 = const()[name = tensor<string, []>("op_6028"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67409536))), lut = tensor<fp16, [4]>([-0x1.07cp-8, -0x1.328p-14, 0x1.fcp-9, 0x1.29p-5]), 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, []>(67819200)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_17_cast = conv(bias = mid_block_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_6028, groups = var_6000, pad = temb_17_pad_0, pad_type = temb_17_pad_type_0, strides = var_6026, 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, []>(67821824)))];
            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, []>(67824448)))];
            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_6038 = const()[name = tensor<string, []>("op_6038"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6040 = const()[name = tensor<string, []>("op_6040"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67827072))), lut = tensor<fp16, [4]>([-0x1.31cp-5, -0x1.b0cp-8, 0x1.a04p-8, 0x1.36cp-5]), 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, []>(71513536)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_119_cast = conv(bias = mid_block_resnets_0_conv2_bias_to_fp16, dilations = var_6040, groups = var_6000, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = var_6038, 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, []>(71516160)))];
            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, []>(71518784)))];
            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_6060 = const()[name = tensor<string, []>("op_6060"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6062 = const()[name = tensor<string, []>("op_6062"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71521408))), lut = tensor<fp16, [4]>([-0x1.75cp-5, -0x1.b6cp-7, 0x1.b6cp-7, 0x1.748p-5]), 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, []>(71931072)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_123_cast = conv(bias = mid_block_attentions_0_proj_in_bias_to_fp16, dilations = var_6062, groups = var_6000, pad = hidden_states_123_pad_0, pad_type = hidden_states_123_pad_type_0, strides = var_6060, 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_6067 = const()[name = tensor<string, []>("op_6067"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<fp16, [2, 1280, 1, 64]> inputs_37_cast = reshape(shape = var_6067, x = hidden_states_123_cast)[name = tensor<string, []>("inputs_37_cast")];
            tensor<int32, [1]> var_6077 = const()[name = tensor<string, []>("op_6077"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_37_cast = reduce_mean(axes = var_6077, keep_dims = var_5995, 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_6081 = const()[name = tensor<string, []>("op_6081"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_6082_cast = reduce_mean(axes = var_6081, keep_dims = var_5995, x = zero_mean_sq_37_cast)[name = tensor<string, []>("op_6082_cast")];
            tensor<fp16, []> var_6083_to_fp16 = const()[name = tensor<string, []>("op_6083_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_6084_cast = add(x = var_6082_cast, y = var_6083_to_fp16)[name = tensor<string, []>("op_6084_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_6084_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_6088_to_fp16 = const()[name = tensor<string, []>("op_6088_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71933696)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6089_cast = add(x = out_37_cast, y = var_6088_to_fp16)[name = tensor<string, []>("op_6089_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6091_to_fp16 = const()[name = tensor<string, []>("op_6091_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71936320)))];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_125_cast = mul(x = var_6089_cast, y = var_6091_to_fp16)[name = tensor<string, []>("hidden_states_125_cast")];
            tensor<int32, [2]> var_6098 = const()[name = tensor<string, []>("op_6098"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6100 = const()[name = tensor<string, []>("op_6100"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71938944))), lut = tensor<fp16, [4]>([-0x1.aa8p-6, -0x1.d68p-8, 0x1.d4cp-8, 0x1.a9p-6]), 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_6100, groups = var_6000, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = var_6098, 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_6104 = const()[name = tensor<string, []>("op_6104"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6106 = const()[name = tensor<string, []>("op_6106"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72348608))), lut = tensor<fp16, [4]>([-0x1.c18p-6, -0x1.f1p-8, 0x1.f48p-8, 0x1.c2cp-6]), 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_6106, groups = var_6000, pad = k_49_pad_0, pad_type = k_49_pad_type_0, strides = var_6104, 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_6110 = const()[name = tensor<string, []>("op_6110"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6112 = const()[name = tensor<string, []>("op_6112"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72758272))), lut = tensor<fp16, [4]>([-0x1.208p-5, -0x1.43cp-7, 0x1.468p-7, 0x1.218p-5]), 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_6112, groups = var_6000, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = var_6110, 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_6116_begin_0 = const()[name = tensor<string, []>("op_6116_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6116_end_0 = const()[name = tensor<string, []>("op_6116_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_6116_end_mask_0 = const()[name = tensor<string, []>("op_6116_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6116_cast = slice_by_index(begin = var_6116_begin_0, end = var_6116_end_0, end_mask = var_6116_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6116_cast")];
            tensor<int32, [4]> var_6120_begin_0 = const()[name = tensor<string, []>("op_6120_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6120_end_0 = const()[name = tensor<string, []>("op_6120_end_0"), val = tensor<int32, [4]>([2, 128, 1, 64])];
            tensor<bool, [4]> var_6120_end_mask_0 = const()[name = tensor<string, []>("op_6120_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6120_cast = slice_by_index(begin = var_6120_begin_0, end = var_6120_end_0, end_mask = var_6120_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6120_cast")];
            tensor<int32, [4]> var_6124_begin_0 = const()[name = tensor<string, []>("op_6124_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6124_end_0 = const()[name = tensor<string, []>("op_6124_end_0"), val = tensor<int32, [4]>([2, 192, 1, 64])];
            tensor<bool, [4]> var_6124_end_mask_0 = const()[name = tensor<string, []>("op_6124_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6124_cast = slice_by_index(begin = var_6124_begin_0, end = var_6124_end_0, end_mask = var_6124_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6124_cast")];
            tensor<int32, [4]> var_6128_begin_0 = const()[name = tensor<string, []>("op_6128_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6128_end_0 = const()[name = tensor<string, []>("op_6128_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_6128_end_mask_0 = const()[name = tensor<string, []>("op_6128_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6128_cast = slice_by_index(begin = var_6128_begin_0, end = var_6128_end_0, end_mask = var_6128_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6128_cast")];
            tensor<int32, [4]> var_6132_begin_0 = const()[name = tensor<string, []>("op_6132_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6132_end_0 = const()[name = tensor<string, []>("op_6132_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_6132_end_mask_0 = const()[name = tensor<string, []>("op_6132_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6132_cast = slice_by_index(begin = var_6132_begin_0, end = var_6132_end_0, end_mask = var_6132_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6132_cast")];
            tensor<int32, [4]> var_6136_begin_0 = const()[name = tensor<string, []>("op_6136_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6136_end_0 = const()[name = tensor<string, []>("op_6136_end_0"), val = tensor<int32, [4]>([2, 384, 1, 64])];
            tensor<bool, [4]> var_6136_end_mask_0 = const()[name = tensor<string, []>("op_6136_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6136_cast = slice_by_index(begin = var_6136_begin_0, end = var_6136_end_0, end_mask = var_6136_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6136_cast")];
            tensor<int32, [4]> var_6140_begin_0 = const()[name = tensor<string, []>("op_6140_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6140_end_0 = const()[name = tensor<string, []>("op_6140_end_0"), val = tensor<int32, [4]>([2, 448, 1, 64])];
            tensor<bool, [4]> var_6140_end_mask_0 = const()[name = tensor<string, []>("op_6140_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6140_cast = slice_by_index(begin = var_6140_begin_0, end = var_6140_end_0, end_mask = var_6140_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6140_cast")];
            tensor<int32, [4]> var_6144_begin_0 = const()[name = tensor<string, []>("op_6144_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6144_end_0 = const()[name = tensor<string, []>("op_6144_end_0"), val = tensor<int32, [4]>([2, 512, 1, 64])];
            tensor<bool, [4]> var_6144_end_mask_0 = const()[name = tensor<string, []>("op_6144_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6144_cast = slice_by_index(begin = var_6144_begin_0, end = var_6144_end_0, end_mask = var_6144_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6144_cast")];
            tensor<int32, [4]> var_6148_begin_0 = const()[name = tensor<string, []>("op_6148_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6148_end_0 = const()[name = tensor<string, []>("op_6148_end_0"), val = tensor<int32, [4]>([2, 576, 1, 64])];
            tensor<bool, [4]> var_6148_end_mask_0 = const()[name = tensor<string, []>("op_6148_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6148_cast = slice_by_index(begin = var_6148_begin_0, end = var_6148_end_0, end_mask = var_6148_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6148_cast")];
            tensor<int32, [4]> var_6152_begin_0 = const()[name = tensor<string, []>("op_6152_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6152_end_0 = const()[name = tensor<string, []>("op_6152_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_6152_end_mask_0 = const()[name = tensor<string, []>("op_6152_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6152_cast = slice_by_index(begin = var_6152_begin_0, end = var_6152_end_0, end_mask = var_6152_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6152_cast")];
            tensor<int32, [4]> var_6156_begin_0 = const()[name = tensor<string, []>("op_6156_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6156_end_0 = const()[name = tensor<string, []>("op_6156_end_0"), val = tensor<int32, [4]>([2, 704, 1, 64])];
            tensor<bool, [4]> var_6156_end_mask_0 = const()[name = tensor<string, []>("op_6156_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6156_cast = slice_by_index(begin = var_6156_begin_0, end = var_6156_end_0, end_mask = var_6156_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6156_cast")];
            tensor<int32, [4]> var_6160_begin_0 = const()[name = tensor<string, []>("op_6160_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6160_end_0 = const()[name = tensor<string, []>("op_6160_end_0"), val = tensor<int32, [4]>([2, 768, 1, 64])];
            tensor<bool, [4]> var_6160_end_mask_0 = const()[name = tensor<string, []>("op_6160_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6160_cast = slice_by_index(begin = var_6160_begin_0, end = var_6160_end_0, end_mask = var_6160_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6160_cast")];
            tensor<int32, [4]> var_6164_begin_0 = const()[name = tensor<string, []>("op_6164_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6164_end_0 = const()[name = tensor<string, []>("op_6164_end_0"), val = tensor<int32, [4]>([2, 832, 1, 64])];
            tensor<bool, [4]> var_6164_end_mask_0 = const()[name = tensor<string, []>("op_6164_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6164_cast = slice_by_index(begin = var_6164_begin_0, end = var_6164_end_0, end_mask = var_6164_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6164_cast")];
            tensor<int32, [4]> var_6168_begin_0 = const()[name = tensor<string, []>("op_6168_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6168_end_0 = const()[name = tensor<string, []>("op_6168_end_0"), val = tensor<int32, [4]>([2, 896, 1, 64])];
            tensor<bool, [4]> var_6168_end_mask_0 = const()[name = tensor<string, []>("op_6168_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6168_cast = slice_by_index(begin = var_6168_begin_0, end = var_6168_end_0, end_mask = var_6168_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6168_cast")];
            tensor<int32, [4]> var_6172_begin_0 = const()[name = tensor<string, []>("op_6172_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6172_end_0 = const()[name = tensor<string, []>("op_6172_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_6172_end_mask_0 = const()[name = tensor<string, []>("op_6172_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6172_cast = slice_by_index(begin = var_6172_begin_0, end = var_6172_end_0, end_mask = var_6172_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6172_cast")];
            tensor<int32, [4]> var_6176_begin_0 = const()[name = tensor<string, []>("op_6176_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6176_end_0 = const()[name = tensor<string, []>("op_6176_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_6176_end_mask_0 = const()[name = tensor<string, []>("op_6176_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6176_cast = slice_by_index(begin = var_6176_begin_0, end = var_6176_end_0, end_mask = var_6176_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6176_cast")];
            tensor<int32, [4]> var_6180_begin_0 = const()[name = tensor<string, []>("op_6180_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6180_end_0 = const()[name = tensor<string, []>("op_6180_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 64])];
            tensor<bool, [4]> var_6180_end_mask_0 = const()[name = tensor<string, []>("op_6180_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6180_cast = slice_by_index(begin = var_6180_begin_0, end = var_6180_end_0, end_mask = var_6180_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6180_cast")];
            tensor<int32, [4]> var_6184_begin_0 = const()[name = tensor<string, []>("op_6184_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6184_end_0 = const()[name = tensor<string, []>("op_6184_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 64])];
            tensor<bool, [4]> var_6184_end_mask_0 = const()[name = tensor<string, []>("op_6184_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6184_cast = slice_by_index(begin = var_6184_begin_0, end = var_6184_end_0, end_mask = var_6184_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6184_cast")];
            tensor<int32, [4]> var_6188_begin_0 = const()[name = tensor<string, []>("op_6188_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6188_end_0 = const()[name = tensor<string, []>("op_6188_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 64])];
            tensor<bool, [4]> var_6188_end_mask_0 = const()[name = tensor<string, []>("op_6188_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6188_cast = slice_by_index(begin = var_6188_begin_0, end = var_6188_end_0, end_mask = var_6188_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6188_cast")];
            tensor<int32, [4]> var_6192_begin_0 = const()[name = tensor<string, []>("op_6192_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6192_end_0 = const()[name = tensor<string, []>("op_6192_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_6192_end_mask_0 = const()[name = tensor<string, []>("op_6192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6192_cast = slice_by_index(begin = var_6192_begin_0, end = var_6192_end_0, end_mask = var_6192_end_mask_0, x = q_25_cast)[name = tensor<string, []>("op_6192_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_6199_begin_0 = const()[name = tensor<string, []>("op_6199_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6199_end_0 = const()[name = tensor<string, []>("op_6199_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_6199_end_mask_0 = const()[name = tensor<string, []>("op_6199_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, 64]> var_6199_cast = slice_by_index(begin = var_6199_begin_0, end = var_6199_end_0, end_mask = var_6199_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6199_cast")];
            tensor<int32, [4]> var_6203_begin_0 = const()[name = tensor<string, []>("op_6203_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6203_end_0 = const()[name = tensor<string, []>("op_6203_end_0"), val = tensor<int32, [4]>([2, 64, 1, 128])];
            tensor<bool, [4]> var_6203_end_mask_0 = const()[name = tensor<string, []>("op_6203_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6203_cast = slice_by_index(begin = var_6203_begin_0, end = var_6203_end_0, end_mask = var_6203_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6203_cast")];
            tensor<int32, [4]> var_6207_begin_0 = const()[name = tensor<string, []>("op_6207_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6207_end_0 = const()[name = tensor<string, []>("op_6207_end_0"), val = tensor<int32, [4]>([2, 64, 1, 192])];
            tensor<bool, [4]> var_6207_end_mask_0 = const()[name = tensor<string, []>("op_6207_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6207_cast = slice_by_index(begin = var_6207_begin_0, end = var_6207_end_0, end_mask = var_6207_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6207_cast")];
            tensor<int32, [4]> var_6211_begin_0 = const()[name = tensor<string, []>("op_6211_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6211_end_0 = const()[name = tensor<string, []>("op_6211_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_6211_end_mask_0 = const()[name = tensor<string, []>("op_6211_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6211_cast = slice_by_index(begin = var_6211_begin_0, end = var_6211_end_0, end_mask = var_6211_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6211_cast")];
            tensor<int32, [4]> var_6215_begin_0 = const()[name = tensor<string, []>("op_6215_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6215_end_0 = const()[name = tensor<string, []>("op_6215_end_0"), val = tensor<int32, [4]>([2, 64, 1, 320])];
            tensor<bool, [4]> var_6215_end_mask_0 = const()[name = tensor<string, []>("op_6215_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6215_cast = slice_by_index(begin = var_6215_begin_0, end = var_6215_end_0, end_mask = var_6215_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6215_cast")];
            tensor<int32, [4]> var_6219_begin_0 = const()[name = tensor<string, []>("op_6219_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6219_end_0 = const()[name = tensor<string, []>("op_6219_end_0"), val = tensor<int32, [4]>([2, 64, 1, 384])];
            tensor<bool, [4]> var_6219_end_mask_0 = const()[name = tensor<string, []>("op_6219_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6219_cast = slice_by_index(begin = var_6219_begin_0, end = var_6219_end_0, end_mask = var_6219_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6219_cast")];
            tensor<int32, [4]> var_6223_begin_0 = const()[name = tensor<string, []>("op_6223_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6223_end_0 = const()[name = tensor<string, []>("op_6223_end_0"), val = tensor<int32, [4]>([2, 64, 1, 448])];
            tensor<bool, [4]> var_6223_end_mask_0 = const()[name = tensor<string, []>("op_6223_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6223_cast = slice_by_index(begin = var_6223_begin_0, end = var_6223_end_0, end_mask = var_6223_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6223_cast")];
            tensor<int32, [4]> var_6227_begin_0 = const()[name = tensor<string, []>("op_6227_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6227_end_0 = const()[name = tensor<string, []>("op_6227_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_6227_end_mask_0 = const()[name = tensor<string, []>("op_6227_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6227_cast = slice_by_index(begin = var_6227_begin_0, end = var_6227_end_0, end_mask = var_6227_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6227_cast")];
            tensor<int32, [4]> var_6231_begin_0 = const()[name = tensor<string, []>("op_6231_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6231_end_0 = const()[name = tensor<string, []>("op_6231_end_0"), val = tensor<int32, [4]>([2, 64, 1, 576])];
            tensor<bool, [4]> var_6231_end_mask_0 = const()[name = tensor<string, []>("op_6231_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6231_cast = slice_by_index(begin = var_6231_begin_0, end = var_6231_end_0, end_mask = var_6231_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6231_cast")];
            tensor<int32, [4]> var_6235_begin_0 = const()[name = tensor<string, []>("op_6235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6235_end_0 = const()[name = tensor<string, []>("op_6235_end_0"), val = tensor<int32, [4]>([2, 64, 1, 640])];
            tensor<bool, [4]> var_6235_end_mask_0 = const()[name = tensor<string, []>("op_6235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6235_cast = slice_by_index(begin = var_6235_begin_0, end = var_6235_end_0, end_mask = var_6235_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6235_cast")];
            tensor<int32, [4]> var_6239_begin_0 = const()[name = tensor<string, []>("op_6239_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6239_end_0 = const()[name = tensor<string, []>("op_6239_end_0"), val = tensor<int32, [4]>([2, 64, 1, 704])];
            tensor<bool, [4]> var_6239_end_mask_0 = const()[name = tensor<string, []>("op_6239_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6239_cast = slice_by_index(begin = var_6239_begin_0, end = var_6239_end_0, end_mask = var_6239_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6239_cast")];
            tensor<int32, [4]> var_6243_begin_0 = const()[name = tensor<string, []>("op_6243_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6243_end_0 = const()[name = tensor<string, []>("op_6243_end_0"), val = tensor<int32, [4]>([2, 64, 1, 768])];
            tensor<bool, [4]> var_6243_end_mask_0 = const()[name = tensor<string, []>("op_6243_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6243_cast = slice_by_index(begin = var_6243_begin_0, end = var_6243_end_0, end_mask = var_6243_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6243_cast")];
            tensor<int32, [4]> var_6247_begin_0 = const()[name = tensor<string, []>("op_6247_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6247_end_0 = const()[name = tensor<string, []>("op_6247_end_0"), val = tensor<int32, [4]>([2, 64, 1, 832])];
            tensor<bool, [4]> var_6247_end_mask_0 = const()[name = tensor<string, []>("op_6247_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6247_cast = slice_by_index(begin = var_6247_begin_0, end = var_6247_end_0, end_mask = var_6247_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6247_cast")];
            tensor<int32, [4]> var_6251_begin_0 = const()[name = tensor<string, []>("op_6251_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6251_end_0 = const()[name = tensor<string, []>("op_6251_end_0"), val = tensor<int32, [4]>([2, 64, 1, 896])];
            tensor<bool, [4]> var_6251_end_mask_0 = const()[name = tensor<string, []>("op_6251_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6251_cast = slice_by_index(begin = var_6251_begin_0, end = var_6251_end_0, end_mask = var_6251_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6251_cast")];
            tensor<int32, [4]> var_6255_begin_0 = const()[name = tensor<string, []>("op_6255_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6255_end_0 = const()[name = tensor<string, []>("op_6255_end_0"), val = tensor<int32, [4]>([2, 64, 1, 960])];
            tensor<bool, [4]> var_6255_end_mask_0 = const()[name = tensor<string, []>("op_6255_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6255_cast = slice_by_index(begin = var_6255_begin_0, end = var_6255_end_0, end_mask = var_6255_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6255_cast")];
            tensor<int32, [4]> var_6259_begin_0 = const()[name = tensor<string, []>("op_6259_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6259_end_0 = const()[name = tensor<string, []>("op_6259_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_6259_end_mask_0 = const()[name = tensor<string, []>("op_6259_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6259_cast = slice_by_index(begin = var_6259_begin_0, end = var_6259_end_0, end_mask = var_6259_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6259_cast")];
            tensor<int32, [4]> var_6263_begin_0 = const()[name = tensor<string, []>("op_6263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_6263_end_0 = const()[name = tensor<string, []>("op_6263_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1088])];
            tensor<bool, [4]> var_6263_end_mask_0 = const()[name = tensor<string, []>("op_6263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6263_cast = slice_by_index(begin = var_6263_begin_0, end = var_6263_end_0, end_mask = var_6263_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6263_cast")];
            tensor<int32, [4]> var_6267_begin_0 = const()[name = tensor<string, []>("op_6267_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_6267_end_0 = const()[name = tensor<string, []>("op_6267_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1152])];
            tensor<bool, [4]> var_6267_end_mask_0 = const()[name = tensor<string, []>("op_6267_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6267_cast = slice_by_index(begin = var_6267_begin_0, end = var_6267_end_0, end_mask = var_6267_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6267_cast")];
            tensor<int32, [4]> var_6271_begin_0 = const()[name = tensor<string, []>("op_6271_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_6271_end_0 = const()[name = tensor<string, []>("op_6271_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1216])];
            tensor<bool, [4]> var_6271_end_mask_0 = const()[name = tensor<string, []>("op_6271_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6271_cast = slice_by_index(begin = var_6271_begin_0, end = var_6271_end_0, end_mask = var_6271_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6271_cast")];
            tensor<int32, [4]> var_6275_begin_0 = const()[name = tensor<string, []>("op_6275_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_6275_end_0 = const()[name = tensor<string, []>("op_6275_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1280])];
            tensor<bool, [4]> var_6275_end_mask_0 = const()[name = tensor<string, []>("op_6275_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 64]> var_6275_cast = slice_by_index(begin = var_6275_begin_0, end = var_6275_end_0, end_mask = var_6275_end_mask_0, x = transpose_19)[name = tensor<string, []>("op_6275_cast")];
            tensor<int32, [4]> var_6277_begin_0 = const()[name = tensor<string, []>("op_6277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6277_end_0 = const()[name = tensor<string, []>("op_6277_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_6277_end_mask_0 = const()[name = tensor<string, []>("op_6277_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6277_cast = slice_by_index(begin = var_6277_begin_0, end = var_6277_end_0, end_mask = var_6277_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6277_cast")];
            tensor<int32, [4]> var_6281_begin_0 = const()[name = tensor<string, []>("op_6281_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6281_end_0 = const()[name = tensor<string, []>("op_6281_end_0"), val = tensor<int32, [4]>([2, 128, 1, 64])];
            tensor<bool, [4]> var_6281_end_mask_0 = const()[name = tensor<string, []>("op_6281_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6281_cast = slice_by_index(begin = var_6281_begin_0, end = var_6281_end_0, end_mask = var_6281_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6281_cast")];
            tensor<int32, [4]> var_6285_begin_0 = const()[name = tensor<string, []>("op_6285_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6285_end_0 = const()[name = tensor<string, []>("op_6285_end_0"), val = tensor<int32, [4]>([2, 192, 1, 64])];
            tensor<bool, [4]> var_6285_end_mask_0 = const()[name = tensor<string, []>("op_6285_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6285_cast = slice_by_index(begin = var_6285_begin_0, end = var_6285_end_0, end_mask = var_6285_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6285_cast")];
            tensor<int32, [4]> var_6289_begin_0 = const()[name = tensor<string, []>("op_6289_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6289_end_0 = const()[name = tensor<string, []>("op_6289_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_6289_end_mask_0 = const()[name = tensor<string, []>("op_6289_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6289_cast = slice_by_index(begin = var_6289_begin_0, end = var_6289_end_0, end_mask = var_6289_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6289_cast")];
            tensor<int32, [4]> var_6293_begin_0 = const()[name = tensor<string, []>("op_6293_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6293_end_0 = const()[name = tensor<string, []>("op_6293_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_6293_end_mask_0 = const()[name = tensor<string, []>("op_6293_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6293_cast = slice_by_index(begin = var_6293_begin_0, end = var_6293_end_0, end_mask = var_6293_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6293_cast")];
            tensor<int32, [4]> var_6297_begin_0 = const()[name = tensor<string, []>("op_6297_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6297_end_0 = const()[name = tensor<string, []>("op_6297_end_0"), val = tensor<int32, [4]>([2, 384, 1, 64])];
            tensor<bool, [4]> var_6297_end_mask_0 = const()[name = tensor<string, []>("op_6297_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6297_cast = slice_by_index(begin = var_6297_begin_0, end = var_6297_end_0, end_mask = var_6297_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6297_cast")];
            tensor<int32, [4]> var_6301_begin_0 = const()[name = tensor<string, []>("op_6301_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6301_end_0 = const()[name = tensor<string, []>("op_6301_end_0"), val = tensor<int32, [4]>([2, 448, 1, 64])];
            tensor<bool, [4]> var_6301_end_mask_0 = const()[name = tensor<string, []>("op_6301_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6301_cast = slice_by_index(begin = var_6301_begin_0, end = var_6301_end_0, end_mask = var_6301_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6301_cast")];
            tensor<int32, [4]> var_6305_begin_0 = const()[name = tensor<string, []>("op_6305_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6305_end_0 = const()[name = tensor<string, []>("op_6305_end_0"), val = tensor<int32, [4]>([2, 512, 1, 64])];
            tensor<bool, [4]> var_6305_end_mask_0 = const()[name = tensor<string, []>("op_6305_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6305_cast = slice_by_index(begin = var_6305_begin_0, end = var_6305_end_0, end_mask = var_6305_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6305_cast")];
            tensor<int32, [4]> var_6309_begin_0 = const()[name = tensor<string, []>("op_6309_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6309_end_0 = const()[name = tensor<string, []>("op_6309_end_0"), val = tensor<int32, [4]>([2, 576, 1, 64])];
            tensor<bool, [4]> var_6309_end_mask_0 = const()[name = tensor<string, []>("op_6309_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6309_cast = slice_by_index(begin = var_6309_begin_0, end = var_6309_end_0, end_mask = var_6309_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6309_cast")];
            tensor<int32, [4]> var_6313_begin_0 = const()[name = tensor<string, []>("op_6313_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6313_end_0 = const()[name = tensor<string, []>("op_6313_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_6313_end_mask_0 = const()[name = tensor<string, []>("op_6313_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6313_cast = slice_by_index(begin = var_6313_begin_0, end = var_6313_end_0, end_mask = var_6313_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6313_cast")];
            tensor<int32, [4]> var_6317_begin_0 = const()[name = tensor<string, []>("op_6317_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6317_end_0 = const()[name = tensor<string, []>("op_6317_end_0"), val = tensor<int32, [4]>([2, 704, 1, 64])];
            tensor<bool, [4]> var_6317_end_mask_0 = const()[name = tensor<string, []>("op_6317_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6317_cast = slice_by_index(begin = var_6317_begin_0, end = var_6317_end_0, end_mask = var_6317_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6317_cast")];
            tensor<int32, [4]> var_6321_begin_0 = const()[name = tensor<string, []>("op_6321_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6321_end_0 = const()[name = tensor<string, []>("op_6321_end_0"), val = tensor<int32, [4]>([2, 768, 1, 64])];
            tensor<bool, [4]> var_6321_end_mask_0 = const()[name = tensor<string, []>("op_6321_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6321_cast = slice_by_index(begin = var_6321_begin_0, end = var_6321_end_0, end_mask = var_6321_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6321_cast")];
            tensor<int32, [4]> var_6325_begin_0 = const()[name = tensor<string, []>("op_6325_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6325_end_0 = const()[name = tensor<string, []>("op_6325_end_0"), val = tensor<int32, [4]>([2, 832, 1, 64])];
            tensor<bool, [4]> var_6325_end_mask_0 = const()[name = tensor<string, []>("op_6325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6325_cast = slice_by_index(begin = var_6325_begin_0, end = var_6325_end_0, end_mask = var_6325_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6325_cast")];
            tensor<int32, [4]> var_6329_begin_0 = const()[name = tensor<string, []>("op_6329_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6329_end_0 = const()[name = tensor<string, []>("op_6329_end_0"), val = tensor<int32, [4]>([2, 896, 1, 64])];
            tensor<bool, [4]> var_6329_end_mask_0 = const()[name = tensor<string, []>("op_6329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6329_cast = slice_by_index(begin = var_6329_begin_0, end = var_6329_end_0, end_mask = var_6329_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6329_cast")];
            tensor<int32, [4]> var_6333_begin_0 = const()[name = tensor<string, []>("op_6333_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6333_end_0 = const()[name = tensor<string, []>("op_6333_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_6333_end_mask_0 = const()[name = tensor<string, []>("op_6333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6333_cast = slice_by_index(begin = var_6333_begin_0, end = var_6333_end_0, end_mask = var_6333_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6333_cast")];
            tensor<int32, [4]> var_6337_begin_0 = const()[name = tensor<string, []>("op_6337_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6337_end_0 = const()[name = tensor<string, []>("op_6337_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_6337_end_mask_0 = const()[name = tensor<string, []>("op_6337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6337_cast = slice_by_index(begin = var_6337_begin_0, end = var_6337_end_0, end_mask = var_6337_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6337_cast")];
            tensor<int32, [4]> var_6341_begin_0 = const()[name = tensor<string, []>("op_6341_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6341_end_0 = const()[name = tensor<string, []>("op_6341_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 64])];
            tensor<bool, [4]> var_6341_end_mask_0 = const()[name = tensor<string, []>("op_6341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6341_cast = slice_by_index(begin = var_6341_begin_0, end = var_6341_end_0, end_mask = var_6341_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6341_cast")];
            tensor<int32, [4]> var_6345_begin_0 = const()[name = tensor<string, []>("op_6345_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6345_end_0 = const()[name = tensor<string, []>("op_6345_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 64])];
            tensor<bool, [4]> var_6345_end_mask_0 = const()[name = tensor<string, []>("op_6345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6345_cast = slice_by_index(begin = var_6345_begin_0, end = var_6345_end_0, end_mask = var_6345_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6345_cast")];
            tensor<int32, [4]> var_6349_begin_0 = const()[name = tensor<string, []>("op_6349_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6349_end_0 = const()[name = tensor<string, []>("op_6349_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 64])];
            tensor<bool, [4]> var_6349_end_mask_0 = const()[name = tensor<string, []>("op_6349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6349_cast = slice_by_index(begin = var_6349_begin_0, end = var_6349_end_0, end_mask = var_6349_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6349_cast")];
            tensor<int32, [4]> var_6353_begin_0 = const()[name = tensor<string, []>("op_6353_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6353_end_0 = const()[name = tensor<string, []>("op_6353_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_6353_end_mask_0 = const()[name = tensor<string, []>("op_6353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6353_cast = slice_by_index(begin = var_6353_begin_0, end = var_6353_end_0, end_mask = var_6353_end_mask_0, x = v_25_cast)[name = tensor<string, []>("op_6353_cast")];
            tensor<string, []> var_6357_equation_0 = const()[name = tensor<string, []>("op_6357_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6357_cast = einsum(equation = var_6357_equation_0, values = (var_6199_cast, var_6116_cast))[name = tensor<string, []>("op_6357_cast")];
            tensor<fp16, []> var_6358_to_fp16 = const()[name = tensor<string, []>("op_6358_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_161_cast = mul(x = var_6357_cast, y = var_6358_to_fp16)[name = tensor<string, []>("aw_161_cast")];
            tensor<string, []> var_6361_equation_0 = const()[name = tensor<string, []>("op_6361_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6361_cast = einsum(equation = var_6361_equation_0, values = (var_6203_cast, var_6120_cast))[name = tensor<string, []>("op_6361_cast")];
            tensor<fp16, []> var_6362_to_fp16 = const()[name = tensor<string, []>("op_6362_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_163_cast = mul(x = var_6361_cast, y = var_6362_to_fp16)[name = tensor<string, []>("aw_163_cast")];
            tensor<string, []> var_6365_equation_0 = const()[name = tensor<string, []>("op_6365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6365_cast = einsum(equation = var_6365_equation_0, values = (var_6207_cast, var_6124_cast))[name = tensor<string, []>("op_6365_cast")];
            tensor<fp16, []> var_6366_to_fp16 = const()[name = tensor<string, []>("op_6366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_165_cast = mul(x = var_6365_cast, y = var_6366_to_fp16)[name = tensor<string, []>("aw_165_cast")];
            tensor<string, []> var_6369_equation_0 = const()[name = tensor<string, []>("op_6369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6369_cast = einsum(equation = var_6369_equation_0, values = (var_6211_cast, var_6128_cast))[name = tensor<string, []>("op_6369_cast")];
            tensor<fp16, []> var_6370_to_fp16 = const()[name = tensor<string, []>("op_6370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_167_cast = mul(x = var_6369_cast, y = var_6370_to_fp16)[name = tensor<string, []>("aw_167_cast")];
            tensor<string, []> var_6373_equation_0 = const()[name = tensor<string, []>("op_6373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6373_cast = einsum(equation = var_6373_equation_0, values = (var_6215_cast, var_6132_cast))[name = tensor<string, []>("op_6373_cast")];
            tensor<fp16, []> var_6374_to_fp16 = const()[name = tensor<string, []>("op_6374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_169_cast = mul(x = var_6373_cast, y = var_6374_to_fp16)[name = tensor<string, []>("aw_169_cast")];
            tensor<string, []> var_6377_equation_0 = const()[name = tensor<string, []>("op_6377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6377_cast = einsum(equation = var_6377_equation_0, values = (var_6219_cast, var_6136_cast))[name = tensor<string, []>("op_6377_cast")];
            tensor<fp16, []> var_6378_to_fp16 = const()[name = tensor<string, []>("op_6378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_171_cast = mul(x = var_6377_cast, y = var_6378_to_fp16)[name = tensor<string, []>("aw_171_cast")];
            tensor<string, []> var_6381_equation_0 = const()[name = tensor<string, []>("op_6381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6381_cast = einsum(equation = var_6381_equation_0, values = (var_6223_cast, var_6140_cast))[name = tensor<string, []>("op_6381_cast")];
            tensor<fp16, []> var_6382_to_fp16 = const()[name = tensor<string, []>("op_6382_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_173_cast = mul(x = var_6381_cast, y = var_6382_to_fp16)[name = tensor<string, []>("aw_173_cast")];
            tensor<string, []> var_6385_equation_0 = const()[name = tensor<string, []>("op_6385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6385_cast = einsum(equation = var_6385_equation_0, values = (var_6227_cast, var_6144_cast))[name = tensor<string, []>("op_6385_cast")];
            tensor<fp16, []> var_6386_to_fp16 = const()[name = tensor<string, []>("op_6386_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_175_cast = mul(x = var_6385_cast, y = var_6386_to_fp16)[name = tensor<string, []>("aw_175_cast")];
            tensor<string, []> var_6389_equation_0 = const()[name = tensor<string, []>("op_6389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6389_cast = einsum(equation = var_6389_equation_0, values = (var_6231_cast, var_6148_cast))[name = tensor<string, []>("op_6389_cast")];
            tensor<fp16, []> var_6390_to_fp16 = const()[name = tensor<string, []>("op_6390_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_177_cast = mul(x = var_6389_cast, y = var_6390_to_fp16)[name = tensor<string, []>("aw_177_cast")];
            tensor<string, []> var_6393_equation_0 = const()[name = tensor<string, []>("op_6393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6393_cast = einsum(equation = var_6393_equation_0, values = (var_6235_cast, var_6152_cast))[name = tensor<string, []>("op_6393_cast")];
            tensor<fp16, []> var_6394_to_fp16 = const()[name = tensor<string, []>("op_6394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_179_cast = mul(x = var_6393_cast, y = var_6394_to_fp16)[name = tensor<string, []>("aw_179_cast")];
            tensor<string, []> var_6397_equation_0 = const()[name = tensor<string, []>("op_6397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6397_cast = einsum(equation = var_6397_equation_0, values = (var_6239_cast, var_6156_cast))[name = tensor<string, []>("op_6397_cast")];
            tensor<fp16, []> var_6398_to_fp16 = const()[name = tensor<string, []>("op_6398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_181_cast = mul(x = var_6397_cast, y = var_6398_to_fp16)[name = tensor<string, []>("aw_181_cast")];
            tensor<string, []> var_6401_equation_0 = const()[name = tensor<string, []>("op_6401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6401_cast = einsum(equation = var_6401_equation_0, values = (var_6243_cast, var_6160_cast))[name = tensor<string, []>("op_6401_cast")];
            tensor<fp16, []> var_6402_to_fp16 = const()[name = tensor<string, []>("op_6402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_183_cast = mul(x = var_6401_cast, y = var_6402_to_fp16)[name = tensor<string, []>("aw_183_cast")];
            tensor<string, []> var_6405_equation_0 = const()[name = tensor<string, []>("op_6405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6405_cast = einsum(equation = var_6405_equation_0, values = (var_6247_cast, var_6164_cast))[name = tensor<string, []>("op_6405_cast")];
            tensor<fp16, []> var_6406_to_fp16 = const()[name = tensor<string, []>("op_6406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_185_cast = mul(x = var_6405_cast, y = var_6406_to_fp16)[name = tensor<string, []>("aw_185_cast")];
            tensor<string, []> var_6409_equation_0 = const()[name = tensor<string, []>("op_6409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6409_cast = einsum(equation = var_6409_equation_0, values = (var_6251_cast, var_6168_cast))[name = tensor<string, []>("op_6409_cast")];
            tensor<fp16, []> var_6410_to_fp16 = const()[name = tensor<string, []>("op_6410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_187_cast = mul(x = var_6409_cast, y = var_6410_to_fp16)[name = tensor<string, []>("aw_187_cast")];
            tensor<string, []> var_6413_equation_0 = const()[name = tensor<string, []>("op_6413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6413_cast = einsum(equation = var_6413_equation_0, values = (var_6255_cast, var_6172_cast))[name = tensor<string, []>("op_6413_cast")];
            tensor<fp16, []> var_6414_to_fp16 = const()[name = tensor<string, []>("op_6414_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_189_cast = mul(x = var_6413_cast, y = var_6414_to_fp16)[name = tensor<string, []>("aw_189_cast")];
            tensor<string, []> var_6417_equation_0 = const()[name = tensor<string, []>("op_6417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6417_cast = einsum(equation = var_6417_equation_0, values = (var_6259_cast, var_6176_cast))[name = tensor<string, []>("op_6417_cast")];
            tensor<fp16, []> var_6418_to_fp16 = const()[name = tensor<string, []>("op_6418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_191_cast = mul(x = var_6417_cast, y = var_6418_to_fp16)[name = tensor<string, []>("aw_191_cast")];
            tensor<string, []> var_6421_equation_0 = const()[name = tensor<string, []>("op_6421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6421_cast = einsum(equation = var_6421_equation_0, values = (var_6263_cast, var_6180_cast))[name = tensor<string, []>("op_6421_cast")];
            tensor<fp16, []> var_6422_to_fp16 = const()[name = tensor<string, []>("op_6422_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_193_cast = mul(x = var_6421_cast, y = var_6422_to_fp16)[name = tensor<string, []>("aw_193_cast")];
            tensor<string, []> var_6425_equation_0 = const()[name = tensor<string, []>("op_6425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6425_cast = einsum(equation = var_6425_equation_0, values = (var_6267_cast, var_6184_cast))[name = tensor<string, []>("op_6425_cast")];
            tensor<fp16, []> var_6426_to_fp16 = const()[name = tensor<string, []>("op_6426_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_195_cast = mul(x = var_6425_cast, y = var_6426_to_fp16)[name = tensor<string, []>("aw_195_cast")];
            tensor<string, []> var_6429_equation_0 = const()[name = tensor<string, []>("op_6429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6429_cast = einsum(equation = var_6429_equation_0, values = (var_6271_cast, var_6188_cast))[name = tensor<string, []>("op_6429_cast")];
            tensor<fp16, []> var_6430_to_fp16 = const()[name = tensor<string, []>("op_6430_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_197_cast = mul(x = var_6429_cast, y = var_6430_to_fp16)[name = tensor<string, []>("aw_197_cast")];
            tensor<string, []> var_6433_equation_0 = const()[name = tensor<string, []>("op_6433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 64, 1, 64]> var_6433_cast = einsum(equation = var_6433_equation_0, values = (var_6275_cast, var_6192_cast))[name = tensor<string, []>("op_6433_cast")];
            tensor<fp16, []> var_6434_to_fp16 = const()[name = tensor<string, []>("op_6434_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 64, 1, 64]> aw_199_cast = mul(x = var_6433_cast, y = var_6434_to_fp16)[name = tensor<string, []>("aw_199_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6436_cast = softmax(axis = var_6000, x = aw_161_cast)[name = tensor<string, []>("op_6436_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6437_cast = softmax(axis = var_6000, x = aw_163_cast)[name = tensor<string, []>("op_6437_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6438_cast = softmax(axis = var_6000, x = aw_165_cast)[name = tensor<string, []>("op_6438_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6439_cast = softmax(axis = var_6000, x = aw_167_cast)[name = tensor<string, []>("op_6439_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6440_cast = softmax(axis = var_6000, x = aw_169_cast)[name = tensor<string, []>("op_6440_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6441_cast = softmax(axis = var_6000, x = aw_171_cast)[name = tensor<string, []>("op_6441_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6442_cast = softmax(axis = var_6000, x = aw_173_cast)[name = tensor<string, []>("op_6442_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6443_cast = softmax(axis = var_6000, x = aw_175_cast)[name = tensor<string, []>("op_6443_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6444_cast = softmax(axis = var_6000, x = aw_177_cast)[name = tensor<string, []>("op_6444_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6445_cast = softmax(axis = var_6000, x = aw_179_cast)[name = tensor<string, []>("op_6445_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6446_cast = softmax(axis = var_6000, x = aw_181_cast)[name = tensor<string, []>("op_6446_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6447_cast = softmax(axis = var_6000, x = aw_183_cast)[name = tensor<string, []>("op_6447_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6448_cast = softmax(axis = var_6000, x = aw_185_cast)[name = tensor<string, []>("op_6448_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6449_cast = softmax(axis = var_6000, x = aw_187_cast)[name = tensor<string, []>("op_6449_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6450_cast = softmax(axis = var_6000, x = aw_189_cast)[name = tensor<string, []>("op_6450_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6451_cast = softmax(axis = var_6000, x = aw_191_cast)[name = tensor<string, []>("op_6451_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6452_cast = softmax(axis = var_6000, x = aw_193_cast)[name = tensor<string, []>("op_6452_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6453_cast = softmax(axis = var_6000, x = aw_195_cast)[name = tensor<string, []>("op_6453_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6454_cast = softmax(axis = var_6000, x = aw_197_cast)[name = tensor<string, []>("op_6454_cast")];
            tensor<fp16, [2, 64, 1, 64]> var_6455_cast = softmax(axis = var_6000, x = aw_199_cast)[name = tensor<string, []>("op_6455_cast")];
            tensor<string, []> var_6457_equation_0 = const()[name = tensor<string, []>("op_6457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6457_cast = einsum(equation = var_6457_equation_0, values = (var_6277_cast, var_6436_cast))[name = tensor<string, []>("op_6457_cast")];
            tensor<string, []> var_6459_equation_0 = const()[name = tensor<string, []>("op_6459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6459_cast = einsum(equation = var_6459_equation_0, values = (var_6281_cast, var_6437_cast))[name = tensor<string, []>("op_6459_cast")];
            tensor<string, []> var_6461_equation_0 = const()[name = tensor<string, []>("op_6461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6461_cast = einsum(equation = var_6461_equation_0, values = (var_6285_cast, var_6438_cast))[name = tensor<string, []>("op_6461_cast")];
            tensor<string, []> var_6463_equation_0 = const()[name = tensor<string, []>("op_6463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6463_cast = einsum(equation = var_6463_equation_0, values = (var_6289_cast, var_6439_cast))[name = tensor<string, []>("op_6463_cast")];
            tensor<string, []> var_6465_equation_0 = const()[name = tensor<string, []>("op_6465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6465_cast = einsum(equation = var_6465_equation_0, values = (var_6293_cast, var_6440_cast))[name = tensor<string, []>("op_6465_cast")];
            tensor<string, []> var_6467_equation_0 = const()[name = tensor<string, []>("op_6467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6467_cast = einsum(equation = var_6467_equation_0, values = (var_6297_cast, var_6441_cast))[name = tensor<string, []>("op_6467_cast")];
            tensor<string, []> var_6469_equation_0 = const()[name = tensor<string, []>("op_6469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6469_cast = einsum(equation = var_6469_equation_0, values = (var_6301_cast, var_6442_cast))[name = tensor<string, []>("op_6469_cast")];
            tensor<string, []> var_6471_equation_0 = const()[name = tensor<string, []>("op_6471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6471_cast = einsum(equation = var_6471_equation_0, values = (var_6305_cast, var_6443_cast))[name = tensor<string, []>("op_6471_cast")];
            tensor<string, []> var_6473_equation_0 = const()[name = tensor<string, []>("op_6473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6473_cast = einsum(equation = var_6473_equation_0, values = (var_6309_cast, var_6444_cast))[name = tensor<string, []>("op_6473_cast")];
            tensor<string, []> var_6475_equation_0 = const()[name = tensor<string, []>("op_6475_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6475_cast = einsum(equation = var_6475_equation_0, values = (var_6313_cast, var_6445_cast))[name = tensor<string, []>("op_6475_cast")];
            tensor<string, []> var_6477_equation_0 = const()[name = tensor<string, []>("op_6477_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6477_cast = einsum(equation = var_6477_equation_0, values = (var_6317_cast, var_6446_cast))[name = tensor<string, []>("op_6477_cast")];
            tensor<string, []> var_6479_equation_0 = const()[name = tensor<string, []>("op_6479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6479_cast = einsum(equation = var_6479_equation_0, values = (var_6321_cast, var_6447_cast))[name = tensor<string, []>("op_6479_cast")];
            tensor<string, []> var_6481_equation_0 = const()[name = tensor<string, []>("op_6481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6481_cast = einsum(equation = var_6481_equation_0, values = (var_6325_cast, var_6448_cast))[name = tensor<string, []>("op_6481_cast")];
            tensor<string, []> var_6483_equation_0 = const()[name = tensor<string, []>("op_6483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6483_cast = einsum(equation = var_6483_equation_0, values = (var_6329_cast, var_6449_cast))[name = tensor<string, []>("op_6483_cast")];
            tensor<string, []> var_6485_equation_0 = const()[name = tensor<string, []>("op_6485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6485_cast = einsum(equation = var_6485_equation_0, values = (var_6333_cast, var_6450_cast))[name = tensor<string, []>("op_6485_cast")];
            tensor<string, []> var_6487_equation_0 = const()[name = tensor<string, []>("op_6487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6487_cast = einsum(equation = var_6487_equation_0, values = (var_6337_cast, var_6451_cast))[name = tensor<string, []>("op_6487_cast")];
            tensor<string, []> var_6489_equation_0 = const()[name = tensor<string, []>("op_6489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6489_cast = einsum(equation = var_6489_equation_0, values = (var_6341_cast, var_6452_cast))[name = tensor<string, []>("op_6489_cast")];
            tensor<string, []> var_6491_equation_0 = const()[name = tensor<string, []>("op_6491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6491_cast = einsum(equation = var_6491_equation_0, values = (var_6345_cast, var_6453_cast))[name = tensor<string, []>("op_6491_cast")];
            tensor<string, []> var_6493_equation_0 = const()[name = tensor<string, []>("op_6493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6493_cast = einsum(equation = var_6493_equation_0, values = (var_6349_cast, var_6454_cast))[name = tensor<string, []>("op_6493_cast")];
            tensor<string, []> var_6495_equation_0 = const()[name = tensor<string, []>("op_6495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6495_cast = einsum(equation = var_6495_equation_0, values = (var_6353_cast, var_6455_cast))[name = tensor<string, []>("op_6495_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_6000, interleave = input_215_interleave_0, values = (var_6457_cast, var_6459_cast, var_6461_cast, var_6463_cast, var_6465_cast, var_6467_cast, var_6469_cast, var_6471_cast, var_6473_cast, var_6475_cast, var_6477_cast, var_6479_cast, var_6481_cast, var_6483_cast, var_6485_cast, var_6487_cast, var_6489_cast, var_6491_cast, var_6493_cast, var_6495_cast))[name = tensor<string, []>("input_215_cast")];
            tensor<int32, [2]> var_6501 = const()[name = tensor<string, []>("op_6501"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6503 = const()[name = tensor<string, []>("op_6503"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6505_pad_type_0 = const()[name = tensor<string, []>("op_6505_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6505_pad_0 = const()[name = tensor<string, []>("op_6505_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73167936))), lut = tensor<fp16, [4]>([-0x1.16cp-5, -0x1.3ecp-7, 0x1.45p-7, 0x1.184p-5]), 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, []>(73577600)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6505_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_6503, groups = var_6000, pad = var_6505_pad_0, pad_type = var_6505_pad_type_0, strides = var_6501, weight = mid_block_attentions_0_transformer_blocks_0_attn1_to_out_0_weight_to_fp16_palettized, x = input_215_cast)[name = tensor<string, []>("op_6505_cast")];
            tensor<fp16, [2, 1280, 1, 64]> inputs_39_cast = add(x = var_6505_cast, y = inputs_37_cast)[name = tensor<string, []>("inputs_39_cast")];
            tensor<int32, [1]> var_6509 = const()[name = tensor<string, []>("op_6509"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_39_cast = reduce_mean(axes = var_6509, keep_dims = var_5995, 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_6513 = const()[name = tensor<string, []>("op_6513"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_6514_cast = reduce_mean(axes = var_6513, keep_dims = var_5995, x = zero_mean_sq_39_cast)[name = tensor<string, []>("op_6514_cast")];
            tensor<fp16, []> var_6515_to_fp16 = const()[name = tensor<string, []>("op_6515_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_6516_cast = add(x = var_6514_cast, y = var_6515_to_fp16)[name = tensor<string, []>("op_6516_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_6516_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_6520_to_fp16 = const()[name = tensor<string, []>("op_6520_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73580224)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6521_cast = add(x = out_39_cast, y = var_6520_to_fp16)[name = tensor<string, []>("op_6521_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6523_to_fp16 = const()[name = tensor<string, []>("op_6523_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73582848)))];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_127_cast = mul(x = var_6521_cast, y = var_6523_to_fp16)[name = tensor<string, []>("hidden_states_127_cast")];
            tensor<int32, [2]> var_6530 = const()[name = tensor<string, []>("op_6530"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6532 = const()[name = tensor<string, []>("op_6532"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73585472))), lut = tensor<fp16, [4]>([-0x1.8a8p-6, -0x1.a3cp-8, 0x1.9e8p-8, 0x1.89p-6]), 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_6532, groups = var_6000, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = var_6530, 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_6536 = const()[name = tensor<string, []>("op_6536"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6538 = const()[name = tensor<string, []>("op_6538"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_53_pad_type_0 = const()[name = tensor<string, []>("k_53_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_53_pad_0 = const()[name = tensor<string, []>("k_53_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73995136))), lut = tensor<fp16, [4]>([-0x1.1a8p-5, -0x1.274p-7, 0x1.2d4p-7, 0x1.1d8p-5]), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_53_cast = conv(dilations = var_6538, groups = var_6000, pad = k_53_pad_0, pad_type = k_53_pad_type_0, strides = var_6536, 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_6542 = const()[name = tensor<string, []>("op_6542"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6544 = const()[name = tensor<string, []>("op_6544"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_27_pad_type_0 = const()[name = tensor<string, []>("v_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_27_pad_0 = const()[name = tensor<string, []>("v_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74322880))), lut = tensor<fp16, [4]>([-0x1.954p-5, -0x1.d04p-7, 0x1.d3cp-7, 0x1.964p-5]), name = tensor<string, []>("mid_block_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_27_cast = conv(dilations = var_6544, groups = var_6000, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = var_6542, 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_6548_begin_0 = const()[name = tensor<string, []>("op_6548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6548_end_0 = const()[name = tensor<string, []>("op_6548_end_0"), val = tensor<int32, [4]>([2, 64, 1, 64])];
            tensor<bool, [4]> var_6548_end_mask_0 = const()[name = tensor<string, []>("op_6548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6548_cast = slice_by_index(begin = var_6548_begin_0, end = var_6548_end_0, end_mask = var_6548_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6548_cast")];
            tensor<int32, [4]> var_6552_begin_0 = const()[name = tensor<string, []>("op_6552_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6552_end_0 = const()[name = tensor<string, []>("op_6552_end_0"), val = tensor<int32, [4]>([2, 128, 1, 64])];
            tensor<bool, [4]> var_6552_end_mask_0 = const()[name = tensor<string, []>("op_6552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6552_cast = slice_by_index(begin = var_6552_begin_0, end = var_6552_end_0, end_mask = var_6552_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6552_cast")];
            tensor<int32, [4]> var_6556_begin_0 = const()[name = tensor<string, []>("op_6556_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6556_end_0 = const()[name = tensor<string, []>("op_6556_end_0"), val = tensor<int32, [4]>([2, 192, 1, 64])];
            tensor<bool, [4]> var_6556_end_mask_0 = const()[name = tensor<string, []>("op_6556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6556_cast = slice_by_index(begin = var_6556_begin_0, end = var_6556_end_0, end_mask = var_6556_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6556_cast")];
            tensor<int32, [4]> var_6560_begin_0 = const()[name = tensor<string, []>("op_6560_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6560_end_0 = const()[name = tensor<string, []>("op_6560_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_6560_end_mask_0 = const()[name = tensor<string, []>("op_6560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6560_cast = slice_by_index(begin = var_6560_begin_0, end = var_6560_end_0, end_mask = var_6560_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6560_cast")];
            tensor<int32, [4]> var_6564_begin_0 = const()[name = tensor<string, []>("op_6564_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_6564_end_0 = const()[name = tensor<string, []>("op_6564_end_0"), val = tensor<int32, [4]>([2, 320, 1, 64])];
            tensor<bool, [4]> var_6564_end_mask_0 = const()[name = tensor<string, []>("op_6564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6564_cast = slice_by_index(begin = var_6564_begin_0, end = var_6564_end_0, end_mask = var_6564_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6564_cast")];
            tensor<int32, [4]> var_6568_begin_0 = const()[name = tensor<string, []>("op_6568_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_6568_end_0 = const()[name = tensor<string, []>("op_6568_end_0"), val = tensor<int32, [4]>([2, 384, 1, 64])];
            tensor<bool, [4]> var_6568_end_mask_0 = const()[name = tensor<string, []>("op_6568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6568_cast = slice_by_index(begin = var_6568_begin_0, end = var_6568_end_0, end_mask = var_6568_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6568_cast")];
            tensor<int32, [4]> var_6572_begin_0 = const()[name = tensor<string, []>("op_6572_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_6572_end_0 = const()[name = tensor<string, []>("op_6572_end_0"), val = tensor<int32, [4]>([2, 448, 1, 64])];
            tensor<bool, [4]> var_6572_end_mask_0 = const()[name = tensor<string, []>("op_6572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6572_cast = slice_by_index(begin = var_6572_begin_0, end = var_6572_end_0, end_mask = var_6572_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6572_cast")];
            tensor<int32, [4]> var_6576_begin_0 = const()[name = tensor<string, []>("op_6576_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_6576_end_0 = const()[name = tensor<string, []>("op_6576_end_0"), val = tensor<int32, [4]>([2, 512, 1, 64])];
            tensor<bool, [4]> var_6576_end_mask_0 = const()[name = tensor<string, []>("op_6576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6576_cast = slice_by_index(begin = var_6576_begin_0, end = var_6576_end_0, end_mask = var_6576_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6576_cast")];
            tensor<int32, [4]> var_6580_begin_0 = const()[name = tensor<string, []>("op_6580_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_6580_end_0 = const()[name = tensor<string, []>("op_6580_end_0"), val = tensor<int32, [4]>([2, 576, 1, 64])];
            tensor<bool, [4]> var_6580_end_mask_0 = const()[name = tensor<string, []>("op_6580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6580_cast = slice_by_index(begin = var_6580_begin_0, end = var_6580_end_0, end_mask = var_6580_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6580_cast")];
            tensor<int32, [4]> var_6584_begin_0 = const()[name = tensor<string, []>("op_6584_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_6584_end_0 = const()[name = tensor<string, []>("op_6584_end_0"), val = tensor<int32, [4]>([2, 640, 1, 64])];
            tensor<bool, [4]> var_6584_end_mask_0 = const()[name = tensor<string, []>("op_6584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6584_cast = slice_by_index(begin = var_6584_begin_0, end = var_6584_end_0, end_mask = var_6584_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6584_cast")];
            tensor<int32, [4]> var_6588_begin_0 = const()[name = tensor<string, []>("op_6588_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_6588_end_0 = const()[name = tensor<string, []>("op_6588_end_0"), val = tensor<int32, [4]>([2, 704, 1, 64])];
            tensor<bool, [4]> var_6588_end_mask_0 = const()[name = tensor<string, []>("op_6588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6588_cast = slice_by_index(begin = var_6588_begin_0, end = var_6588_end_0, end_mask = var_6588_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6588_cast")];
            tensor<int32, [4]> var_6592_begin_0 = const()[name = tensor<string, []>("op_6592_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6592_end_0 = const()[name = tensor<string, []>("op_6592_end_0"), val = tensor<int32, [4]>([2, 768, 1, 64])];
            tensor<bool, [4]> var_6592_end_mask_0 = const()[name = tensor<string, []>("op_6592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6592_cast = slice_by_index(begin = var_6592_begin_0, end = var_6592_end_0, end_mask = var_6592_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6592_cast")];
            tensor<int32, [4]> var_6596_begin_0 = const()[name = tensor<string, []>("op_6596_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6596_end_0 = const()[name = tensor<string, []>("op_6596_end_0"), val = tensor<int32, [4]>([2, 832, 1, 64])];
            tensor<bool, [4]> var_6596_end_mask_0 = const()[name = tensor<string, []>("op_6596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6596_cast = slice_by_index(begin = var_6596_begin_0, end = var_6596_end_0, end_mask = var_6596_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6596_cast")];
            tensor<int32, [4]> var_6600_begin_0 = const()[name = tensor<string, []>("op_6600_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6600_end_0 = const()[name = tensor<string, []>("op_6600_end_0"), val = tensor<int32, [4]>([2, 896, 1, 64])];
            tensor<bool, [4]> var_6600_end_mask_0 = const()[name = tensor<string, []>("op_6600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6600_cast = slice_by_index(begin = var_6600_begin_0, end = var_6600_end_0, end_mask = var_6600_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6600_cast")];
            tensor<int32, [4]> var_6604_begin_0 = const()[name = tensor<string, []>("op_6604_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6604_end_0 = const()[name = tensor<string, []>("op_6604_end_0"), val = tensor<int32, [4]>([2, 960, 1, 64])];
            tensor<bool, [4]> var_6604_end_mask_0 = const()[name = tensor<string, []>("op_6604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6604_cast = slice_by_index(begin = var_6604_begin_0, end = var_6604_end_0, end_mask = var_6604_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6604_cast")];
            tensor<int32, [4]> var_6608_begin_0 = const()[name = tensor<string, []>("op_6608_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6608_end_0 = const()[name = tensor<string, []>("op_6608_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_6608_end_mask_0 = const()[name = tensor<string, []>("op_6608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6608_cast = slice_by_index(begin = var_6608_begin_0, end = var_6608_end_0, end_mask = var_6608_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6608_cast")];
            tensor<int32, [4]> var_6612_begin_0 = const()[name = tensor<string, []>("op_6612_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6612_end_0 = const()[name = tensor<string, []>("op_6612_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 64])];
            tensor<bool, [4]> var_6612_end_mask_0 = const()[name = tensor<string, []>("op_6612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6612_cast = slice_by_index(begin = var_6612_begin_0, end = var_6612_end_0, end_mask = var_6612_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6612_cast")];
            tensor<int32, [4]> var_6616_begin_0 = const()[name = tensor<string, []>("op_6616_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6616_end_0 = const()[name = tensor<string, []>("op_6616_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 64])];
            tensor<bool, [4]> var_6616_end_mask_0 = const()[name = tensor<string, []>("op_6616_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6616_cast = slice_by_index(begin = var_6616_begin_0, end = var_6616_end_0, end_mask = var_6616_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6616_cast")];
            tensor<int32, [4]> var_6620_begin_0 = const()[name = tensor<string, []>("op_6620_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6620_end_0 = const()[name = tensor<string, []>("op_6620_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 64])];
            tensor<bool, [4]> var_6620_end_mask_0 = const()[name = tensor<string, []>("op_6620_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6620_cast = slice_by_index(begin = var_6620_begin_0, end = var_6620_end_0, end_mask = var_6620_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6620_cast")];
            tensor<int32, [4]> var_6624_begin_0 = const()[name = tensor<string, []>("op_6624_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6624_end_0 = const()[name = tensor<string, []>("op_6624_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 64])];
            tensor<bool, [4]> var_6624_end_mask_0 = const()[name = tensor<string, []>("op_6624_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 64]> var_6624_cast = slice_by_index(begin = var_6624_begin_0, end = var_6624_end_0, end_mask = var_6624_end_mask_0, x = q_27_cast)[name = tensor<string, []>("op_6624_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_6631_begin_0 = const()[name = tensor<string, []>("op_6631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6631_end_0 = const()[name = tensor<string, []>("op_6631_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_6631_end_mask_0 = const()[name = tensor<string, []>("op_6631_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, 64]> var_6631_cast = slice_by_index(begin = var_6631_begin_0, end = var_6631_end_0, end_mask = var_6631_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6631_cast")];
            tensor<int32, [4]> var_6635_begin_0 = const()[name = tensor<string, []>("op_6635_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_6635_end_0 = const()[name = tensor<string, []>("op_6635_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_6635_end_mask_0 = const()[name = tensor<string, []>("op_6635_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6635_cast = slice_by_index(begin = var_6635_begin_0, end = var_6635_end_0, end_mask = var_6635_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6635_cast")];
            tensor<int32, [4]> var_6639_begin_0 = const()[name = tensor<string, []>("op_6639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_6639_end_0 = const()[name = tensor<string, []>("op_6639_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_6639_end_mask_0 = const()[name = tensor<string, []>("op_6639_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6639_cast = slice_by_index(begin = var_6639_begin_0, end = var_6639_end_0, end_mask = var_6639_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6639_cast")];
            tensor<int32, [4]> var_6643_begin_0 = const()[name = tensor<string, []>("op_6643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_6643_end_0 = const()[name = tensor<string, []>("op_6643_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_6643_end_mask_0 = const()[name = tensor<string, []>("op_6643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6643_cast = slice_by_index(begin = var_6643_begin_0, end = var_6643_end_0, end_mask = var_6643_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6643_cast")];
            tensor<int32, [4]> var_6647_begin_0 = const()[name = tensor<string, []>("op_6647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_6647_end_0 = const()[name = tensor<string, []>("op_6647_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_6647_end_mask_0 = const()[name = tensor<string, []>("op_6647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6647_cast = slice_by_index(begin = var_6647_begin_0, end = var_6647_end_0, end_mask = var_6647_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6647_cast")];
            tensor<int32, [4]> var_6651_begin_0 = const()[name = tensor<string, []>("op_6651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_6651_end_0 = const()[name = tensor<string, []>("op_6651_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_6651_end_mask_0 = const()[name = tensor<string, []>("op_6651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6651_cast = slice_by_index(begin = var_6651_begin_0, end = var_6651_end_0, end_mask = var_6651_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6651_cast")];
            tensor<int32, [4]> var_6655_begin_0 = const()[name = tensor<string, []>("op_6655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_6655_end_0 = const()[name = tensor<string, []>("op_6655_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_6655_end_mask_0 = const()[name = tensor<string, []>("op_6655_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6655_cast = slice_by_index(begin = var_6655_begin_0, end = var_6655_end_0, end_mask = var_6655_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6655_cast")];
            tensor<int32, [4]> var_6659_begin_0 = const()[name = tensor<string, []>("op_6659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_6659_end_0 = const()[name = tensor<string, []>("op_6659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_6659_end_mask_0 = const()[name = tensor<string, []>("op_6659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6659_cast = slice_by_index(begin = var_6659_begin_0, end = var_6659_end_0, end_mask = var_6659_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6659_cast")];
            tensor<int32, [4]> var_6663_begin_0 = const()[name = tensor<string, []>("op_6663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_6663_end_0 = const()[name = tensor<string, []>("op_6663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_6663_end_mask_0 = const()[name = tensor<string, []>("op_6663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6663_cast = slice_by_index(begin = var_6663_begin_0, end = var_6663_end_0, end_mask = var_6663_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6663_cast")];
            tensor<int32, [4]> var_6667_begin_0 = const()[name = tensor<string, []>("op_6667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_6667_end_0 = const()[name = tensor<string, []>("op_6667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_6667_end_mask_0 = const()[name = tensor<string, []>("op_6667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6667_cast = slice_by_index(begin = var_6667_begin_0, end = var_6667_end_0, end_mask = var_6667_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6667_cast")];
            tensor<int32, [4]> var_6671_begin_0 = const()[name = tensor<string, []>("op_6671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_6671_end_0 = const()[name = tensor<string, []>("op_6671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_6671_end_mask_0 = const()[name = tensor<string, []>("op_6671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6671_cast = slice_by_index(begin = var_6671_begin_0, end = var_6671_end_0, end_mask = var_6671_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6671_cast")];
            tensor<int32, [4]> var_6675_begin_0 = const()[name = tensor<string, []>("op_6675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_6675_end_0 = const()[name = tensor<string, []>("op_6675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_6675_end_mask_0 = const()[name = tensor<string, []>("op_6675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6675_cast = slice_by_index(begin = var_6675_begin_0, end = var_6675_end_0, end_mask = var_6675_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6675_cast")];
            tensor<int32, [4]> var_6679_begin_0 = const()[name = tensor<string, []>("op_6679_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_6679_end_0 = const()[name = tensor<string, []>("op_6679_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_6679_end_mask_0 = const()[name = tensor<string, []>("op_6679_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6679_cast = slice_by_index(begin = var_6679_begin_0, end = var_6679_end_0, end_mask = var_6679_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6679_cast")];
            tensor<int32, [4]> var_6683_begin_0 = const()[name = tensor<string, []>("op_6683_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_6683_end_0 = const()[name = tensor<string, []>("op_6683_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_6683_end_mask_0 = const()[name = tensor<string, []>("op_6683_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6683_cast = slice_by_index(begin = var_6683_begin_0, end = var_6683_end_0, end_mask = var_6683_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6683_cast")];
            tensor<int32, [4]> var_6687_begin_0 = const()[name = tensor<string, []>("op_6687_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_6687_end_0 = const()[name = tensor<string, []>("op_6687_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_6687_end_mask_0 = const()[name = tensor<string, []>("op_6687_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_6687_cast = slice_by_index(begin = var_6687_begin_0, end = var_6687_end_0, end_mask = var_6687_end_mask_0, x = transpose_18)[name = tensor<string, []>("op_6687_cast")];
            tensor<int32, [4]> var_6691_begin_0 = const()[name = tensor<string, []>("op_6691_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_6691_end_0 = const()[name = tensor<string, []>("op_6691_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            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, 77, 1, 64]> 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_18)[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, 1024])];
            tensor<int32, [4]> var_6695_end_0 = const()[name = tensor<string, []>("op_6695_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            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, 77, 1, 64]> 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_18)[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, 1088])];
            tensor<int32, [4]> var_6699_end_0 = const()[name = tensor<string, []>("op_6699_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            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, 77, 1, 64]> 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_18)[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, 1152])];
            tensor<int32, [4]> var_6703_end_0 = const()[name = tensor<string, []>("op_6703_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            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, 77, 1, 64]> 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_18)[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, 1216])];
            tensor<int32, [4]> var_6707_end_0 = const()[name = tensor<string, []>("op_6707_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            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, 77, 1, 64]> 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_18)[name = tensor<string, []>("op_6707_cast")];
            tensor<int32, [4]> var_6709_begin_0 = const()[name = tensor<string, []>("op_6709_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_6709_end_0 = const()[name = tensor<string, []>("op_6709_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_6709_end_mask_0 = const()[name = tensor<string, []>("op_6709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6709_cast = slice_by_index(begin = var_6709_begin_0, end = var_6709_end_0, end_mask = var_6709_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6709_cast")];
            tensor<int32, [4]> var_6713_begin_0 = const()[name = tensor<string, []>("op_6713_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_6713_end_0 = const()[name = tensor<string, []>("op_6713_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_6713_end_mask_0 = const()[name = tensor<string, []>("op_6713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6713_cast = slice_by_index(begin = var_6713_begin_0, end = var_6713_end_0, end_mask = var_6713_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6713_cast")];
            tensor<int32, [4]> var_6717_begin_0 = const()[name = tensor<string, []>("op_6717_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_6717_end_0 = const()[name = tensor<string, []>("op_6717_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_6717_end_mask_0 = const()[name = tensor<string, []>("op_6717_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6717_cast = slice_by_index(begin = var_6717_begin_0, end = var_6717_end_0, end_mask = var_6717_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6717_cast")];
            tensor<int32, [4]> var_6721_begin_0 = const()[name = tensor<string, []>("op_6721_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_6721_end_0 = const()[name = tensor<string, []>("op_6721_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 256, 0, 0])];
            tensor<int32, [4]> var_6725_end_0 = const()[name = tensor<string, []>("op_6725_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 384, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 384, 0, 0])];
            tensor<int32, [4]> var_6733_end_0 = const()[name = tensor<string, []>("op_6733_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 448, 0, 0])];
            tensor<int32, [4]> var_6737_end_0 = const()[name = tensor<string, []>("op_6737_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 512, 0, 0])];
            tensor<int32, [4]> var_6741_end_0 = const()[name = tensor<string, []>("op_6741_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 576, 0, 0])];
            tensor<int32, [4]> var_6745_end_0 = const()[name = tensor<string, []>("op_6745_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            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, 64, 1, 77]> 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_27_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, 640, 0, 0])];
            tensor<int32, [4]> var_6749_end_0 = const()[name = tensor<string, []>("op_6749_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            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, 64, 1, 77]> 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_27_cast)[name = tensor<string, []>("op_6749_cast")];
            tensor<int32, [4]> var_6753_begin_0 = const()[name = tensor<string, []>("op_6753_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_6753_end_0 = const()[name = tensor<string, []>("op_6753_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_6753_end_mask_0 = const()[name = tensor<string, []>("op_6753_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6753_cast = slice_by_index(begin = var_6753_begin_0, end = var_6753_end_0, end_mask = var_6753_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6753_cast")];
            tensor<int32, [4]> var_6757_begin_0 = const()[name = tensor<string, []>("op_6757_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_6757_end_0 = const()[name = tensor<string, []>("op_6757_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_6757_end_mask_0 = const()[name = tensor<string, []>("op_6757_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6757_cast = slice_by_index(begin = var_6757_begin_0, end = var_6757_end_0, end_mask = var_6757_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6757_cast")];
            tensor<int32, [4]> var_6761_begin_0 = const()[name = tensor<string, []>("op_6761_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_6761_end_0 = const()[name = tensor<string, []>("op_6761_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_6761_end_mask_0 = const()[name = tensor<string, []>("op_6761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6761_cast = slice_by_index(begin = var_6761_begin_0, end = var_6761_end_0, end_mask = var_6761_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6761_cast")];
            tensor<int32, [4]> var_6765_begin_0 = const()[name = tensor<string, []>("op_6765_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_6765_end_0 = const()[name = tensor<string, []>("op_6765_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_6765_end_mask_0 = const()[name = tensor<string, []>("op_6765_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6765_cast = slice_by_index(begin = var_6765_begin_0, end = var_6765_end_0, end_mask = var_6765_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6765_cast")];
            tensor<int32, [4]> var_6769_begin_0 = const()[name = tensor<string, []>("op_6769_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_6769_end_0 = const()[name = tensor<string, []>("op_6769_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_6769_end_mask_0 = const()[name = tensor<string, []>("op_6769_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6769_cast = slice_by_index(begin = var_6769_begin_0, end = var_6769_end_0, end_mask = var_6769_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6769_cast")];
            tensor<int32, [4]> var_6773_begin_0 = const()[name = tensor<string, []>("op_6773_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_6773_end_0 = const()[name = tensor<string, []>("op_6773_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_6773_end_mask_0 = const()[name = tensor<string, []>("op_6773_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6773_cast = slice_by_index(begin = var_6773_begin_0, end = var_6773_end_0, end_mask = var_6773_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6773_cast")];
            tensor<int32, [4]> var_6777_begin_0 = const()[name = tensor<string, []>("op_6777_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_6777_end_0 = const()[name = tensor<string, []>("op_6777_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_6777_end_mask_0 = const()[name = tensor<string, []>("op_6777_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6777_cast = slice_by_index(begin = var_6777_begin_0, end = var_6777_end_0, end_mask = var_6777_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6777_cast")];
            tensor<int32, [4]> var_6781_begin_0 = const()[name = tensor<string, []>("op_6781_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_6781_end_0 = const()[name = tensor<string, []>("op_6781_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_6781_end_mask_0 = const()[name = tensor<string, []>("op_6781_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6781_cast = slice_by_index(begin = var_6781_begin_0, end = var_6781_end_0, end_mask = var_6781_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6781_cast")];
            tensor<int32, [4]> var_6785_begin_0 = const()[name = tensor<string, []>("op_6785_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_6785_end_0 = const()[name = tensor<string, []>("op_6785_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_6785_end_mask_0 = const()[name = tensor<string, []>("op_6785_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_6785_cast = slice_by_index(begin = var_6785_begin_0, end = var_6785_end_0, end_mask = var_6785_end_mask_0, x = v_27_cast)[name = tensor<string, []>("op_6785_cast")];
            tensor<string, []> var_6789_equation_0 = const()[name = tensor<string, []>("op_6789_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6789_cast = einsum(equation = var_6789_equation_0, values = (var_6631_cast, var_6548_cast))[name = tensor<string, []>("op_6789_cast")];
            tensor<fp16, []> var_6790_to_fp16 = const()[name = tensor<string, []>("op_6790_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_201_cast = mul(x = var_6789_cast, y = var_6790_to_fp16)[name = tensor<string, []>("aw_201_cast")];
            tensor<string, []> var_6793_equation_0 = const()[name = tensor<string, []>("op_6793_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6793_cast = einsum(equation = var_6793_equation_0, values = (var_6635_cast, var_6552_cast))[name = tensor<string, []>("op_6793_cast")];
            tensor<fp16, []> var_6794_to_fp16 = const()[name = tensor<string, []>("op_6794_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_203_cast = mul(x = var_6793_cast, y = var_6794_to_fp16)[name = tensor<string, []>("aw_203_cast")];
            tensor<string, []> var_6797_equation_0 = const()[name = tensor<string, []>("op_6797_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6797_cast = einsum(equation = var_6797_equation_0, values = (var_6639_cast, var_6556_cast))[name = tensor<string, []>("op_6797_cast")];
            tensor<fp16, []> var_6798_to_fp16 = const()[name = tensor<string, []>("op_6798_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_205_cast = mul(x = var_6797_cast, y = var_6798_to_fp16)[name = tensor<string, []>("aw_205_cast")];
            tensor<string, []> var_6801_equation_0 = const()[name = tensor<string, []>("op_6801_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6801_cast = einsum(equation = var_6801_equation_0, values = (var_6643_cast, var_6560_cast))[name = tensor<string, []>("op_6801_cast")];
            tensor<fp16, []> var_6802_to_fp16 = const()[name = tensor<string, []>("op_6802_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_207_cast = mul(x = var_6801_cast, y = var_6802_to_fp16)[name = tensor<string, []>("aw_207_cast")];
            tensor<string, []> var_6805_equation_0 = const()[name = tensor<string, []>("op_6805_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6805_cast = einsum(equation = var_6805_equation_0, values = (var_6647_cast, var_6564_cast))[name = tensor<string, []>("op_6805_cast")];
            tensor<fp16, []> var_6806_to_fp16 = const()[name = tensor<string, []>("op_6806_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_209_cast = mul(x = var_6805_cast, y = var_6806_to_fp16)[name = tensor<string, []>("aw_209_cast")];
            tensor<string, []> var_6809_equation_0 = const()[name = tensor<string, []>("op_6809_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6809_cast = einsum(equation = var_6809_equation_0, values = (var_6651_cast, var_6568_cast))[name = tensor<string, []>("op_6809_cast")];
            tensor<fp16, []> var_6810_to_fp16 = const()[name = tensor<string, []>("op_6810_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_211_cast = mul(x = var_6809_cast, y = var_6810_to_fp16)[name = tensor<string, []>("aw_211_cast")];
            tensor<string, []> var_6813_equation_0 = const()[name = tensor<string, []>("op_6813_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6813_cast = einsum(equation = var_6813_equation_0, values = (var_6655_cast, var_6572_cast))[name = tensor<string, []>("op_6813_cast")];
            tensor<fp16, []> var_6814_to_fp16 = const()[name = tensor<string, []>("op_6814_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_213_cast = mul(x = var_6813_cast, y = var_6814_to_fp16)[name = tensor<string, []>("aw_213_cast")];
            tensor<string, []> var_6817_equation_0 = const()[name = tensor<string, []>("op_6817_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6817_cast = einsum(equation = var_6817_equation_0, values = (var_6659_cast, var_6576_cast))[name = tensor<string, []>("op_6817_cast")];
            tensor<fp16, []> var_6818_to_fp16 = const()[name = tensor<string, []>("op_6818_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_215_cast = mul(x = var_6817_cast, y = var_6818_to_fp16)[name = tensor<string, []>("aw_215_cast")];
            tensor<string, []> var_6821_equation_0 = const()[name = tensor<string, []>("op_6821_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6821_cast = einsum(equation = var_6821_equation_0, values = (var_6663_cast, var_6580_cast))[name = tensor<string, []>("op_6821_cast")];
            tensor<fp16, []> var_6822_to_fp16 = const()[name = tensor<string, []>("op_6822_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_217_cast = mul(x = var_6821_cast, y = var_6822_to_fp16)[name = tensor<string, []>("aw_217_cast")];
            tensor<string, []> var_6825_equation_0 = const()[name = tensor<string, []>("op_6825_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6825_cast = einsum(equation = var_6825_equation_0, values = (var_6667_cast, var_6584_cast))[name = tensor<string, []>("op_6825_cast")];
            tensor<fp16, []> var_6826_to_fp16 = const()[name = tensor<string, []>("op_6826_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_219_cast = mul(x = var_6825_cast, y = var_6826_to_fp16)[name = tensor<string, []>("aw_219_cast")];
            tensor<string, []> var_6829_equation_0 = const()[name = tensor<string, []>("op_6829_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6829_cast = einsum(equation = var_6829_equation_0, values = (var_6671_cast, var_6588_cast))[name = tensor<string, []>("op_6829_cast")];
            tensor<fp16, []> var_6830_to_fp16 = const()[name = tensor<string, []>("op_6830_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_221_cast = mul(x = var_6829_cast, y = var_6830_to_fp16)[name = tensor<string, []>("aw_221_cast")];
            tensor<string, []> var_6833_equation_0 = const()[name = tensor<string, []>("op_6833_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6833_cast = einsum(equation = var_6833_equation_0, values = (var_6675_cast, var_6592_cast))[name = tensor<string, []>("op_6833_cast")];
            tensor<fp16, []> var_6834_to_fp16 = const()[name = tensor<string, []>("op_6834_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_223_cast = mul(x = var_6833_cast, y = var_6834_to_fp16)[name = tensor<string, []>("aw_223_cast")];
            tensor<string, []> var_6837_equation_0 = const()[name = tensor<string, []>("op_6837_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6837_cast = einsum(equation = var_6837_equation_0, values = (var_6679_cast, var_6596_cast))[name = tensor<string, []>("op_6837_cast")];
            tensor<fp16, []> var_6838_to_fp16 = const()[name = tensor<string, []>("op_6838_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_225_cast = mul(x = var_6837_cast, y = var_6838_to_fp16)[name = tensor<string, []>("aw_225_cast")];
            tensor<string, []> var_6841_equation_0 = const()[name = tensor<string, []>("op_6841_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6841_cast = einsum(equation = var_6841_equation_0, values = (var_6683_cast, var_6600_cast))[name = tensor<string, []>("op_6841_cast")];
            tensor<fp16, []> var_6842_to_fp16 = const()[name = tensor<string, []>("op_6842_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_227_cast = mul(x = var_6841_cast, y = var_6842_to_fp16)[name = tensor<string, []>("aw_227_cast")];
            tensor<string, []> var_6845_equation_0 = const()[name = tensor<string, []>("op_6845_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6845_cast = einsum(equation = var_6845_equation_0, values = (var_6687_cast, var_6604_cast))[name = tensor<string, []>("op_6845_cast")];
            tensor<fp16, []> var_6846_to_fp16 = const()[name = tensor<string, []>("op_6846_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_229_cast = mul(x = var_6845_cast, y = var_6846_to_fp16)[name = tensor<string, []>("aw_229_cast")];
            tensor<string, []> var_6849_equation_0 = const()[name = tensor<string, []>("op_6849_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6849_cast = einsum(equation = var_6849_equation_0, values = (var_6691_cast, var_6608_cast))[name = tensor<string, []>("op_6849_cast")];
            tensor<fp16, []> var_6850_to_fp16 = const()[name = tensor<string, []>("op_6850_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_231_cast = mul(x = var_6849_cast, y = var_6850_to_fp16)[name = tensor<string, []>("aw_231_cast")];
            tensor<string, []> var_6853_equation_0 = const()[name = tensor<string, []>("op_6853_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6853_cast = einsum(equation = var_6853_equation_0, values = (var_6695_cast, var_6612_cast))[name = tensor<string, []>("op_6853_cast")];
            tensor<fp16, []> var_6854_to_fp16 = const()[name = tensor<string, []>("op_6854_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_233_cast = mul(x = var_6853_cast, y = var_6854_to_fp16)[name = tensor<string, []>("aw_233_cast")];
            tensor<string, []> var_6857_equation_0 = const()[name = tensor<string, []>("op_6857_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6857_cast = einsum(equation = var_6857_equation_0, values = (var_6699_cast, var_6616_cast))[name = tensor<string, []>("op_6857_cast")];
            tensor<fp16, []> var_6858_to_fp16 = const()[name = tensor<string, []>("op_6858_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_235_cast = mul(x = var_6857_cast, y = var_6858_to_fp16)[name = tensor<string, []>("aw_235_cast")];
            tensor<string, []> var_6861_equation_0 = const()[name = tensor<string, []>("op_6861_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6861_cast = einsum(equation = var_6861_equation_0, values = (var_6703_cast, var_6620_cast))[name = tensor<string, []>("op_6861_cast")];
            tensor<fp16, []> var_6862_to_fp16 = const()[name = tensor<string, []>("op_6862_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_237_cast = mul(x = var_6861_cast, y = var_6862_to_fp16)[name = tensor<string, []>("aw_237_cast")];
            tensor<string, []> var_6865_equation_0 = const()[name = tensor<string, []>("op_6865_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 64]> var_6865_cast = einsum(equation = var_6865_equation_0, values = (var_6707_cast, var_6624_cast))[name = tensor<string, []>("op_6865_cast")];
            tensor<fp16, []> var_6866_to_fp16 = const()[name = tensor<string, []>("op_6866_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 64]> aw_239_cast = mul(x = var_6865_cast, y = var_6866_to_fp16)[name = tensor<string, []>("aw_239_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6868_cast = softmax(axis = var_6000, x = aw_201_cast)[name = tensor<string, []>("op_6868_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6869_cast = softmax(axis = var_6000, x = aw_203_cast)[name = tensor<string, []>("op_6869_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6870_cast = softmax(axis = var_6000, x = aw_205_cast)[name = tensor<string, []>("op_6870_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6871_cast = softmax(axis = var_6000, x = aw_207_cast)[name = tensor<string, []>("op_6871_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6872_cast = softmax(axis = var_6000, x = aw_209_cast)[name = tensor<string, []>("op_6872_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6873_cast = softmax(axis = var_6000, x = aw_211_cast)[name = tensor<string, []>("op_6873_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6874_cast = softmax(axis = var_6000, x = aw_213_cast)[name = tensor<string, []>("op_6874_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6875_cast = softmax(axis = var_6000, x = aw_215_cast)[name = tensor<string, []>("op_6875_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6876_cast = softmax(axis = var_6000, x = aw_217_cast)[name = tensor<string, []>("op_6876_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6877_cast = softmax(axis = var_6000, x = aw_219_cast)[name = tensor<string, []>("op_6877_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6878_cast = softmax(axis = var_6000, x = aw_221_cast)[name = tensor<string, []>("op_6878_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6879_cast = softmax(axis = var_6000, x = aw_223_cast)[name = tensor<string, []>("op_6879_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6880_cast = softmax(axis = var_6000, x = aw_225_cast)[name = tensor<string, []>("op_6880_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6881_cast = softmax(axis = var_6000, x = aw_227_cast)[name = tensor<string, []>("op_6881_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6882_cast = softmax(axis = var_6000, x = aw_229_cast)[name = tensor<string, []>("op_6882_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6883_cast = softmax(axis = var_6000, x = aw_231_cast)[name = tensor<string, []>("op_6883_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6884_cast = softmax(axis = var_6000, x = aw_233_cast)[name = tensor<string, []>("op_6884_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6885_cast = softmax(axis = var_6000, x = aw_235_cast)[name = tensor<string, []>("op_6885_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6886_cast = softmax(axis = var_6000, x = aw_237_cast)[name = tensor<string, []>("op_6886_cast")];
            tensor<fp16, [2, 77, 1, 64]> var_6887_cast = softmax(axis = var_6000, x = aw_239_cast)[name = tensor<string, []>("op_6887_cast")];
            tensor<string, []> var_6889_equation_0 = const()[name = tensor<string, []>("op_6889_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6889_cast = einsum(equation = var_6889_equation_0, values = (var_6709_cast, var_6868_cast))[name = tensor<string, []>("op_6889_cast")];
            tensor<string, []> var_6891_equation_0 = const()[name = tensor<string, []>("op_6891_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6891_cast = einsum(equation = var_6891_equation_0, values = (var_6713_cast, var_6869_cast))[name = tensor<string, []>("op_6891_cast")];
            tensor<string, []> var_6893_equation_0 = const()[name = tensor<string, []>("op_6893_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6893_cast = einsum(equation = var_6893_equation_0, values = (var_6717_cast, var_6870_cast))[name = tensor<string, []>("op_6893_cast")];
            tensor<string, []> var_6895_equation_0 = const()[name = tensor<string, []>("op_6895_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6895_cast = einsum(equation = var_6895_equation_0, values = (var_6721_cast, var_6871_cast))[name = tensor<string, []>("op_6895_cast")];
            tensor<string, []> var_6897_equation_0 = const()[name = tensor<string, []>("op_6897_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6897_cast = einsum(equation = var_6897_equation_0, values = (var_6725_cast, var_6872_cast))[name = tensor<string, []>("op_6897_cast")];
            tensor<string, []> var_6899_equation_0 = const()[name = tensor<string, []>("op_6899_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6899_cast = einsum(equation = var_6899_equation_0, values = (var_6729_cast, var_6873_cast))[name = tensor<string, []>("op_6899_cast")];
            tensor<string, []> var_6901_equation_0 = const()[name = tensor<string, []>("op_6901_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6901_cast = einsum(equation = var_6901_equation_0, values = (var_6733_cast, var_6874_cast))[name = tensor<string, []>("op_6901_cast")];
            tensor<string, []> var_6903_equation_0 = const()[name = tensor<string, []>("op_6903_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6903_cast = einsum(equation = var_6903_equation_0, values = (var_6737_cast, var_6875_cast))[name = tensor<string, []>("op_6903_cast")];
            tensor<string, []> var_6905_equation_0 = const()[name = tensor<string, []>("op_6905_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6905_cast = einsum(equation = var_6905_equation_0, values = (var_6741_cast, var_6876_cast))[name = tensor<string, []>("op_6905_cast")];
            tensor<string, []> var_6907_equation_0 = const()[name = tensor<string, []>("op_6907_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6907_cast = einsum(equation = var_6907_equation_0, values = (var_6745_cast, var_6877_cast))[name = tensor<string, []>("op_6907_cast")];
            tensor<string, []> var_6909_equation_0 = const()[name = tensor<string, []>("op_6909_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6909_cast = einsum(equation = var_6909_equation_0, values = (var_6749_cast, var_6878_cast))[name = tensor<string, []>("op_6909_cast")];
            tensor<string, []> var_6911_equation_0 = const()[name = tensor<string, []>("op_6911_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6911_cast = einsum(equation = var_6911_equation_0, values = (var_6753_cast, var_6879_cast))[name = tensor<string, []>("op_6911_cast")];
            tensor<string, []> var_6913_equation_0 = const()[name = tensor<string, []>("op_6913_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6913_cast = einsum(equation = var_6913_equation_0, values = (var_6757_cast, var_6880_cast))[name = tensor<string, []>("op_6913_cast")];
            tensor<string, []> var_6915_equation_0 = const()[name = tensor<string, []>("op_6915_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6915_cast = einsum(equation = var_6915_equation_0, values = (var_6761_cast, var_6881_cast))[name = tensor<string, []>("op_6915_cast")];
            tensor<string, []> var_6917_equation_0 = const()[name = tensor<string, []>("op_6917_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6917_cast = einsum(equation = var_6917_equation_0, values = (var_6765_cast, var_6882_cast))[name = tensor<string, []>("op_6917_cast")];
            tensor<string, []> var_6919_equation_0 = const()[name = tensor<string, []>("op_6919_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6919_cast = einsum(equation = var_6919_equation_0, values = (var_6769_cast, var_6883_cast))[name = tensor<string, []>("op_6919_cast")];
            tensor<string, []> var_6921_equation_0 = const()[name = tensor<string, []>("op_6921_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6921_cast = einsum(equation = var_6921_equation_0, values = (var_6773_cast, var_6884_cast))[name = tensor<string, []>("op_6921_cast")];
            tensor<string, []> var_6923_equation_0 = const()[name = tensor<string, []>("op_6923_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6923_cast = einsum(equation = var_6923_equation_0, values = (var_6777_cast, var_6885_cast))[name = tensor<string, []>("op_6923_cast")];
            tensor<string, []> var_6925_equation_0 = const()[name = tensor<string, []>("op_6925_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6925_cast = einsum(equation = var_6925_equation_0, values = (var_6781_cast, var_6886_cast))[name = tensor<string, []>("op_6925_cast")];
            tensor<string, []> var_6927_equation_0 = const()[name = tensor<string, []>("op_6927_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 64]> var_6927_cast = einsum(equation = var_6927_equation_0, values = (var_6785_cast, var_6887_cast))[name = tensor<string, []>("op_6927_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_6000, interleave = input_217_interleave_0, values = (var_6889_cast, var_6891_cast, var_6893_cast, var_6895_cast, var_6897_cast, var_6899_cast, var_6901_cast, var_6903_cast, var_6905_cast, var_6907_cast, var_6909_cast, var_6911_cast, var_6913_cast, var_6915_cast, var_6917_cast, var_6919_cast, var_6921_cast, var_6923_cast, var_6925_cast, var_6927_cast))[name = tensor<string, []>("input_217_cast")];
            tensor<int32, [2]> var_6933 = const()[name = tensor<string, []>("op_6933"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6935 = const()[name = tensor<string, []>("op_6935"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6937_pad_type_0 = const()[name = tensor<string, []>("op_6937_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6937_pad_0 = const()[name = tensor<string, []>("op_6937_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74650624))), lut = tensor<fp16, [4]>([-0x1.edcp-6, -0x1.1ccp-7, 0x1.1ccp-7, 0x1.ee4p-6]), 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, []>(75060288)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6937_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_6935, groups = var_6000, pad = var_6937_pad_0, pad_type = var_6937_pad_type_0, strides = var_6933, weight = mid_block_attentions_0_transformer_blocks_0_attn2_to_out_0_weight_to_fp16_palettized, x = input_217_cast)[name = tensor<string, []>("op_6937_cast")];
            tensor<fp16, [2, 1280, 1, 64]> inputs_41_cast = add(x = var_6937_cast, y = inputs_39_cast)[name = tensor<string, []>("inputs_41_cast")];
            tensor<int32, [1]> var_6941 = const()[name = tensor<string, []>("op_6941"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> channels_mean_41_cast = reduce_mean(axes = var_6941, keep_dims = var_5995, 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_6945 = const()[name = tensor<string, []>("op_6945"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 64]> var_6946_cast = reduce_mean(axes = var_6945, keep_dims = var_5995, x = zero_mean_sq_41_cast)[name = tensor<string, []>("op_6946_cast")];
            tensor<fp16, []> var_6947_to_fp16 = const()[name = tensor<string, []>("op_6947_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 64]> var_6948_cast = add(x = var_6946_cast, y = var_6947_to_fp16)[name = tensor<string, []>("op_6948_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_6948_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_6952_to_fp16 = const()[name = tensor<string, []>("op_6952_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75062912)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6953_cast = add(x = out_41_cast, y = var_6952_to_fp16)[name = tensor<string, []>("op_6953_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_6955_to_fp16 = const()[name = tensor<string, []>("op_6955_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75065536)))];
            tensor<fp16, [2, 1280, 1, 64]> input_219_cast = mul(x = var_6953_cast, y = var_6955_to_fp16)[name = tensor<string, []>("input_219_cast")];
            tensor<int32, [2]> var_6963 = const()[name = tensor<string, []>("op_6963"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6965 = const()[name = tensor<string, []>("op_6965"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6967_pad_type_0 = const()[name = tensor<string, []>("op_6967_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6967_pad_0 = const()[name = tensor<string, []>("op_6967_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, [3276800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75068160))), lut = tensor<fp16, [4]>([-0x1.67p-5, -0x1.a5cp-9, 0x1.a64p-9, 0x1.67p-5]), 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, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78345024))), lut = tensor<fp16, [4]>([-0x1.d64p-4, -0x1.52cp-5, 0x1.6a8p-11, 0x1.528p-5]), 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_6967_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_6965, groups = var_6000, pad = var_6967_pad_0, pad_type = var_6967_pad_type_0, strides = var_6963, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_0_proj_weight_to_fp16_palettized, x = input_219_cast)[name = tensor<string, []>("op_6967_cast")];
            tensor<int32, [2]> var_6968_split_sizes_0 = const()[name = tensor<string, []>("op_6968_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_6968_axis_0 = const()[name = tensor<string, []>("op_6968_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 64]> var_6968_cast_0, tensor<fp16, [2, 5120, 1, 64]> var_6968_cast_1 = split(axis = var_6968_axis_0, split_sizes = var_6968_split_sizes_0, x = var_6967_cast)[name = tensor<string, []>("op_6968_cast")];
            tensor<string, []> var_6970_mode_0 = const()[name = tensor<string, []>("op_6970_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 64]> var_6970_cast = gelu(mode = var_6970_mode_0, x = var_6968_cast_1)[name = tensor<string, []>("op_6970_cast")];
            tensor<fp16, [2, 5120, 1, 64]> input_221_cast = mul(x = var_6968_cast_0, y = var_6970_cast)[name = tensor<string, []>("input_221_cast")];
            tensor<int32, [2]> var_6974 = const()[name = tensor<string, []>("op_6974"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6976 = const()[name = tensor<string, []>("op_6976"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_6978_pad_type_0 = const()[name = tensor<string, []>("op_6978_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_6978_pad_0 = const()[name = tensor<string, []>("op_6978_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, [1638400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78347648))), lut = tensor<fp16, [4]>([-0x1.47cp-5, -0x1.2d8p-13, 0x1.7a8p-6, 0x1.dbp-5]), 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, []>(79986112)))];
            tensor<fp16, [2, 1280, 1, 64]> var_6978_cast = conv(bias = mid_block_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_6976, groups = var_6000, pad = var_6978_pad_0, pad_type = var_6978_pad_type_0, strides = var_6974, weight = mid_block_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_221_cast)[name = tensor<string, []>("op_6978_cast")];
            tensor<fp16, [2, 1280, 1, 64]> hidden_states_131_cast = add(x = var_6978_cast, y = inputs_41_cast)[name = tensor<string, []>("hidden_states_131_cast")];
            tensor<int32, [4]> var_6980 = const()[name = tensor<string, []>("op_6980"), val = tensor<int32, [4]>([2, 1280, 8, 8])];
            tensor<fp16, [2, 1280, 8, 8]> input_223_cast = reshape(shape = var_6980, x = hidden_states_131_cast)[name = tensor<string, []>("input_223_cast")];
            tensor<int32, [2]> var_6984 = const()[name = tensor<string, []>("op_6984"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_6986 = const()[name = tensor<string, []>("op_6986"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79988736))), lut = tensor<fp16, [4]>([-0x1.284p-5, -0x1.51cp-7, 0x1.5d8p-7, 0x1.2c4p-5]), 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, []>(80398400)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_133_cast = conv(bias = mid_block_attentions_0_proj_out_bias_to_fp16, dilations = var_6986, groups = var_6000, pad = hidden_states_133_pad_0, pad_type = hidden_states_133_pad_type_0, strides = var_6984, 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, []>(80401024)))];
            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, []>(80403648)))];
            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_7001 = const()[name = tensor<string, []>("op_7001"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7003 = const()[name = tensor<string, []>("op_7003"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80406272))), lut = tensor<fp16, [4]>([-0x1.054p-5, -0x1.584p-8, 0x1.7f4p-8, 0x1.0dp-5]), 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, []>(84092736)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_135_cast = conv(bias = mid_block_resnets_1_conv1_bias_to_fp16, dilations = var_7003, groups = var_6000, pad = hidden_states_135_pad_0, pad_type = hidden_states_135_pad_type_0, strides = var_7001, 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_7009 = const()[name = tensor<string, []>("op_7009"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7011 = const()[name = tensor<string, []>("op_7011"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84095360))), lut = tensor<fp16, [4]>([-0x1.004p-8, -0x1.5acp-14, 0x1.eb4p-9, 0x1.b6p-6]), 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, []>(84505024)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_19_cast = conv(bias = mid_block_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_7011, groups = var_6000, pad = temb_19_pad_0, pad_type = temb_19_pad_type_0, strides = var_7009, 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, []>(84507648)))];
            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, []>(84510272)))];
            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_7021 = const()[name = tensor<string, []>("op_7021"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7023 = const()[name = tensor<string, []>("op_7023"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84512896))), lut = tensor<fp16, [4]>([-0x1.014p-5, -0x1.384p-8, 0x1.4bcp-8, 0x1p-5]), 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, []>(88199360)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_137_cast = conv(bias = mid_block_resnets_1_conv2_bias_to_fp16, dilations = var_7023, groups = var_6000, pad = hidden_states_137_pad_0, pad_type = hidden_states_137_pad_type_0, strides = var_7021, 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_7034 = const()[name = tensor<string, []>("op_7034"), 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_7034, 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88201984))), lut = tensor<fp16, [4]>([0x0p+0, 0x0p+0, 0x0p+0, 0x0p+0]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88202688))), lut = tensor<fp16, [4]>([0x1p+0, 0x1p+0, 0x1p+0, 0x1p+0]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88203392))), lut = tensor<fp16, [4]>([0x1.e1p-4, 0x1.50cp-2, 0x1.774p-3, 0x1.d0cp-2]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88204096))), lut = tensor<fp16, [4]>([-0x1.9c8p-7, -0x1.ba4p-4, -0x1.5e8p-5, -0x1.db8p-3]), 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_7057 = const()[name = tensor<string, []>("op_7057"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7059 = const()[name = tensor<string, []>("op_7059"), 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, [7372800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88204800))), lut = tensor<fp16, [4]>([-0x1.c2p-6, -0x1.e9cp-9, 0x1.7p-8, 0x1.f04p-6]), 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, []>(95577664)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_141_cast = conv(bias = up_blocks_0_resnets_0_conv1_bias_to_fp16, dilations = var_7059, groups = var_7034, pad = hidden_states_141_pad_0, pad_type = hidden_states_141_pad_type_0, strides = var_7057, 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_7065 = const()[name = tensor<string, []>("op_7065"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7067 = const()[name = tensor<string, []>("op_7067"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95580288))), lut = tensor<fp16, [4]>([-0x1.02cp-8, -0x1.a4cp-14, 0x1.ebp-9, 0x1.f64p-6]), 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, []>(95989952)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_21_cast = conv(bias = up_blocks_0_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_7067, groups = var_7034, pad = temb_21_pad_0, pad_type = temb_21_pad_type_0, strides = var_7065, 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, []>(95992576)))];
            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, []>(95995200)))];
            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_7077 = const()[name = tensor<string, []>("op_7077"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7079 = const()[name = tensor<string, []>("op_7079"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95997824))), lut = tensor<fp16, [4]>([-0x1.488p-5, -0x1.fccp-11, 0x1.848p-6, 0x1.d54p-4]), 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, []>(99684288)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_143_cast = conv(bias = up_blocks_0_resnets_0_conv2_bias_to_fp16, dilations = var_7079, groups = var_7034, pad = hidden_states_143_pad_0, pad_type = hidden_states_143_pad_type_0, strides = var_7077, 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_7084 = const()[name = tensor<string, []>("op_7084"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7086 = const()[name = tensor<string, []>("op_7086"), 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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99686912))), lut = tensor<fp16, [4]>([-0x1.3d8p-6, -0x1.37p-8, 0x1.3a4p-8, 0x1.3ecp-6]), 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, []>(100506176)))];
            tensor<fp16, [2, 1280, 8, 8]> x_5_cast = conv(bias = up_blocks_0_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_7086, groups = var_7034, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = var_7084, 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_7034, 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100508800))), lut = tensor<fp16, [4]>([0x1.8c8p-3, 0x1.70cp-2, 0x1.0ecp-1, 0x1.1fcp-2]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100509504))), lut = tensor<fp16, [4]>([-0x1.88p-6, -0x1.794p-2, -0x1.0e4p-3, -0x1.314p-1]), 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_7104 = const()[name = tensor<string, []>("op_7104"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7106 = const()[name = tensor<string, []>("op_7106"), 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, [7372800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100510208))), lut = tensor<fp16, [4]>([-0x1.03p-5, -0x1.1b8p-8, 0x1.34p-7, 0x1.408p-5]), 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, []>(107883072)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_147_cast = conv(bias = up_blocks_0_resnets_1_conv1_bias_to_fp16, dilations = var_7106, groups = var_7034, pad = hidden_states_147_pad_0, pad_type = hidden_states_147_pad_type_0, strides = var_7104, 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_7112 = const()[name = tensor<string, []>("op_7112"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7114 = const()[name = tensor<string, []>("op_7114"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107885696))), lut = tensor<fp16, [4]>([-0x1.108p-8, -0x1.4f8p-13, 0x1.f78p-9, 0x1.208p-5]), 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, []>(108295360)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_23_cast = conv(bias = up_blocks_0_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_7114, groups = var_7034, pad = temb_23_pad_0, pad_type = temb_23_pad_type_0, strides = var_7112, 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, []>(108297984)))];
            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, []>(108300608)))];
            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_7124 = const()[name = tensor<string, []>("op_7124"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7126 = const()[name = tensor<string, []>("op_7126"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108303232))), lut = tensor<fp16, [4]>([-0x1.9ecp-5, -0x1.9c8p-9, 0x1.43cp-6, 0x1.438p-4]), 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, []>(111989696)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_149_cast = conv(bias = up_blocks_0_resnets_1_conv2_bias_to_fp16, dilations = var_7126, groups = var_7034, pad = hidden_states_149_pad_0, pad_type = hidden_states_149_pad_type_0, strides = var_7124, 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_7131 = const()[name = tensor<string, []>("op_7131"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7133 = const()[name = tensor<string, []>("op_7133"), 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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111992320))), lut = tensor<fp16, [4]>([-0x1.9bcp-7, -0x1.8a8p-9, 0x1.93p-9, 0x1.a08p-7]), 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, []>(112811584)))];
            tensor<fp16, [2, 1280, 8, 8]> x_7_cast = conv(bias = up_blocks_0_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_7133, groups = var_7034, pad = x_7_pad_0, pad_type = x_7_pad_type_0, strides = var_7131, 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_7034, 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112814208))), lut = tensor<fp16, [4]>([0x1.3c8p-2, 0x1.c08p-2, 0x1.a64p-3, 0x1.484p-1]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112814912))), lut = tensor<fp16, [4]>([-0x1.0bcp-3, -0x1.02cp-6, -0x1.4e4p-1, -0x1.648p-2]), 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_7151 = const()[name = tensor<string, []>("op_7151"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7153 = const()[name = tensor<string, []>("op_7153"), 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, [7372800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112815616))), lut = tensor<fp16, [4]>([-0x1.964p-5, -0x1.c3cp-7, 0x1.f94p-9, 0x1.27cp-5]), 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, []>(120188480)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_153_cast = conv(bias = up_blocks_0_resnets_2_conv1_bias_to_fp16, dilations = var_7153, groups = var_7034, pad = hidden_states_153_pad_0, pad_type = hidden_states_153_pad_type_0, strides = var_7151, 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_7159 = const()[name = tensor<string, []>("op_7159"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7161 = const()[name = tensor<string, []>("op_7161"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120191104))), lut = tensor<fp16, [4]>([-0x1.16cp-5, -0x1.724p-9, 0x1.7b4p-9, 0x1.3e8p-5]), 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, []>(120600768)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_25_cast = conv(bias = up_blocks_0_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_7161, groups = var_7034, pad = temb_25_pad_0, pad_type = temb_25_pad_type_0, strides = var_7159, 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, []>(120603392)))];
            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, []>(120606016)))];
            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_7171 = const()[name = tensor<string, []>("op_7171"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7173 = const()[name = tensor<string, []>("op_7173"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120608640))), lut = tensor<fp16, [4]>([-0x1.424p-5, -0x1.384p-9, 0x1.4fp-6, 0x1.16cp-4]), 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, []>(124295104)))];
            tensor<fp16, [2, 1280, 8, 8]> hidden_states_155_cast = conv(bias = up_blocks_0_resnets_2_conv2_bias_to_fp16, dilations = var_7173, groups = var_7034, pad = hidden_states_155_pad_0, pad_type = hidden_states_155_pad_type_0, strides = var_7171, 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_7178 = const()[name = tensor<string, []>("op_7178"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7180 = const()[name = tensor<string, []>("op_7180"), 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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124297728))), lut = tensor<fp16, [4]>([-0x1.a6cp-8, -0x1.e68p-10, 0x1.e44p-10, 0x1.a5cp-8]), 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, []>(125116992)))];
            tensor<fp16, [2, 1280, 8, 8]> x_9_cast = conv(bias = up_blocks_0_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_7180, groups = var_7034, pad = x_9_pad_0, pad_type = x_9_pad_type_0, strides = var_7178, 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_7189 = const()[name = tensor<string, []>("op_7189"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7191 = const()[name = tensor<string, []>("op_7191"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125119616))), lut = tensor<fp16, [4]>([-0x1.b84p-7, -0x1.004p-8, 0x1.f64p-9, 0x1.b44p-7]), 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, []>(128806080)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_157_cast = conv(bias = up_blocks_0_upsamplers_0_conv_bias_to_fp16, dilations = var_7191, groups = var_7034, pad = hidden_states_157_pad_0, pad_type = hidden_states_157_pad_type_0, strides = var_7189, 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_7223 = const()[name = tensor<string, []>("op_7223"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_7228 = const()[name = tensor<string, []>("op_7228"), 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_7228, 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128808704))), lut = tensor<fp16, [4]>([0x1.438p-2, 0x1.6ep-8, 0x1.378p-1, 0x1.afp-2]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128809408))), lut = tensor<fp16, [4]>([-0x1.a18p-6, -0x1.288p-1, -0x1.434p+0, -0x1.f98p-3]), 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_7257 = const()[name = tensor<string, []>("op_7257"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7259 = const()[name = tensor<string, []>("op_7259"), 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, [7372800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128810112))), lut = tensor<fp16, [4]>([-0x1.44p-4, -0x1.eb4p-6, 0x1.07p-8, 0x1.b98p-5]), 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, []>(136182976)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_159_cast = conv(bias = up_blocks_1_resnets_0_conv1_bias_to_fp16, dilations = var_7259, groups = var_7228, pad = hidden_states_159_pad_0, pad_type = hidden_states_159_pad_type_0, strides = var_7257, 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_7265 = const()[name = tensor<string, []>("op_7265"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7267 = const()[name = tensor<string, []>("op_7267"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136185600))), lut = tensor<fp16, [4]>([-0x1.588p+1, -0x1.ff8p-9, 0x1.85cp-9, 0x1.a4p-5]), 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, []>(136595264)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_27_cast = conv(bias = up_blocks_1_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_7267, groups = var_7228, pad = temb_27_pad_0, pad_type = temb_27_pad_type_0, strides = var_7265, 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, []>(136597888)))];
            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, []>(136600512)))];
            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_7277 = const()[name = tensor<string, []>("op_7277"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7279 = const()[name = tensor<string, []>("op_7279"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136603136))), lut = tensor<fp16, [4]>([-0x1.1f4p-4, -0x1.47cp-6, 0x1.47cp-6, 0x1.1f4p-4]), 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, []>(140289600)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_161_cast = conv(bias = up_blocks_1_resnets_0_conv2_bias_to_fp16, dilations = var_7279, groups = var_7228, pad = hidden_states_161_pad_0, pad_type = hidden_states_161_pad_type_0, strides = var_7277, 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_7284 = const()[name = tensor<string, []>("op_7284"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7286 = const()[name = tensor<string, []>("op_7286"), 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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140292224))), lut = tensor<fp16, [4]>([-0x1.1b8p-5, -0x1.2b8p-10, 0x1.484p-6, 0x1.87cp-5]), 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, []>(141111488)))];
            tensor<fp16, [2, 1280, 16, 16]> x_11_cast = conv(bias = up_blocks_1_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_7286, groups = var_7228, pad = x_11_pad_0, pad_type = x_11_pad_type_0, strides = var_7284, 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, []>(141114112)))];
            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, []>(141116736)))];
            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_7306 = const()[name = tensor<string, []>("op_7306"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7308 = const()[name = tensor<string, []>("op_7308"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141119360))), lut = tensor<fp16, [4]>([-0x1.8ecp-4, -0x1.db8p-6, 0x1.e28p-6, 0x1.908p-4]), 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, []>(141529024)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_165_cast = conv(bias = up_blocks_1_attentions_0_proj_in_bias_to_fp16, dilations = var_7308, groups = var_7228, pad = hidden_states_165_pad_0, pad_type = hidden_states_165_pad_type_0, strides = var_7306, 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_7313 = const()[name = tensor<string, []>("op_7313"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_43_cast = reshape(shape = var_7313, x = hidden_states_165_cast)[name = tensor<string, []>("inputs_43_cast")];
            tensor<int32, [1]> var_7323 = const()[name = tensor<string, []>("op_7323"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_43_cast = reduce_mean(axes = var_7323, keep_dims = var_7223, 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_7327 = const()[name = tensor<string, []>("op_7327"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7328_cast = reduce_mean(axes = var_7327, keep_dims = var_7223, x = zero_mean_sq_43_cast)[name = tensor<string, []>("op_7328_cast")];
            tensor<fp16, []> var_7329_to_fp16 = const()[name = tensor<string, []>("op_7329_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7330_cast = add(x = var_7328_cast, y = var_7329_to_fp16)[name = tensor<string, []>("op_7330_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_7330_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_7334_to_fp16 = const()[name = tensor<string, []>("op_7334_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141531648)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7335_cast = add(x = out_43_cast, y = var_7334_to_fp16)[name = tensor<string, []>("op_7335_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7337_to_fp16 = const()[name = tensor<string, []>("op_7337_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141534272)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_167_cast = mul(x = var_7335_cast, y = var_7337_to_fp16)[name = tensor<string, []>("hidden_states_167_cast")];
            tensor<int32, [2]> var_7344 = const()[name = tensor<string, []>("op_7344"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7346 = const()[name = tensor<string, []>("op_7346"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141536896))), lut = tensor<fp16, [4]>([-0x1.75p-4, -0x1.b84p-6, 0x1.b6cp-6, 0x1.754p-4]), 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_7346, groups = var_7228, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = var_7344, 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_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, []> 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141946560))), lut = tensor<fp16, [4]>([-0x1.75p-4, -0x1.b74p-6, 0x1.b6cp-6, 0x1.754p-4]), 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_7352, groups = var_7228, pad = k_57_pad_0, pad_type = k_57_pad_type_0, strides = var_7350, 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_7356 = const()[name = tensor<string, []>("op_7356"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7358 = const()[name = tensor<string, []>("op_7358"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142356224))), lut = tensor<fp16, [4]>([-0x1.56p-4, -0x1.964p-6, 0x1.94cp-6, 0x1.558p-4]), 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_7358, groups = var_7228, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = var_7356, 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_7362_begin_0 = const()[name = tensor<string, []>("op_7362_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7362_end_0 = const()[name = tensor<string, []>("op_7362_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_7362_end_mask_0 = const()[name = tensor<string, []>("op_7362_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7362_cast = slice_by_index(begin = var_7362_begin_0, end = var_7362_end_0, end_mask = var_7362_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7362_cast")];
            tensor<int32, [4]> var_7366_begin_0 = const()[name = tensor<string, []>("op_7366_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7366_end_0 = const()[name = tensor<string, []>("op_7366_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_7366_end_mask_0 = const()[name = tensor<string, []>("op_7366_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7366_cast = slice_by_index(begin = var_7366_begin_0, end = var_7366_end_0, end_mask = var_7366_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7366_cast")];
            tensor<int32, [4]> var_7370_begin_0 = const()[name = tensor<string, []>("op_7370_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7370_end_0 = const()[name = tensor<string, []>("op_7370_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_7370_end_mask_0 = const()[name = tensor<string, []>("op_7370_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7370_cast = slice_by_index(begin = var_7370_begin_0, end = var_7370_end_0, end_mask = var_7370_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7370_cast")];
            tensor<int32, [4]> var_7374_begin_0 = const()[name = tensor<string, []>("op_7374_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7374_end_0 = const()[name = tensor<string, []>("op_7374_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7374_end_mask_0 = const()[name = tensor<string, []>("op_7374_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7374_cast = slice_by_index(begin = var_7374_begin_0, end = var_7374_end_0, end_mask = var_7374_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7374_cast")];
            tensor<int32, [4]> var_7378_begin_0 = const()[name = tensor<string, []>("op_7378_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7378_end_0 = const()[name = tensor<string, []>("op_7378_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7378_end_mask_0 = const()[name = tensor<string, []>("op_7378_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7378_cast = slice_by_index(begin = var_7378_begin_0, end = var_7378_end_0, end_mask = var_7378_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7378_cast")];
            tensor<int32, [4]> var_7382_begin_0 = const()[name = tensor<string, []>("op_7382_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7382_end_0 = const()[name = tensor<string, []>("op_7382_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_7382_end_mask_0 = const()[name = tensor<string, []>("op_7382_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7382_cast = slice_by_index(begin = var_7382_begin_0, end = var_7382_end_0, end_mask = var_7382_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7382_cast")];
            tensor<int32, [4]> var_7386_begin_0 = const()[name = tensor<string, []>("op_7386_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7386_end_0 = const()[name = tensor<string, []>("op_7386_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_7386_end_mask_0 = const()[name = tensor<string, []>("op_7386_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7386_cast = slice_by_index(begin = var_7386_begin_0, end = var_7386_end_0, end_mask = var_7386_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7386_cast")];
            tensor<int32, [4]> var_7390_begin_0 = const()[name = tensor<string, []>("op_7390_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7390_end_0 = const()[name = tensor<string, []>("op_7390_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_7390_end_mask_0 = const()[name = tensor<string, []>("op_7390_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7390_cast = slice_by_index(begin = var_7390_begin_0, end = var_7390_end_0, end_mask = var_7390_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7390_cast")];
            tensor<int32, [4]> var_7394_begin_0 = const()[name = tensor<string, []>("op_7394_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7394_end_0 = const()[name = tensor<string, []>("op_7394_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_7394_end_mask_0 = const()[name = tensor<string, []>("op_7394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7394_cast = slice_by_index(begin = var_7394_begin_0, end = var_7394_end_0, end_mask = var_7394_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7394_cast")];
            tensor<int32, [4]> var_7398_begin_0 = const()[name = tensor<string, []>("op_7398_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7398_end_0 = const()[name = tensor<string, []>("op_7398_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7398_end_mask_0 = const()[name = tensor<string, []>("op_7398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7398_cast = slice_by_index(begin = var_7398_begin_0, end = var_7398_end_0, end_mask = var_7398_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7398_cast")];
            tensor<int32, [4]> var_7402_begin_0 = const()[name = tensor<string, []>("op_7402_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7402_end_0 = const()[name = tensor<string, []>("op_7402_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_7402_end_mask_0 = const()[name = tensor<string, []>("op_7402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7402_cast = slice_by_index(begin = var_7402_begin_0, end = var_7402_end_0, end_mask = var_7402_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7402_cast")];
            tensor<int32, [4]> var_7406_begin_0 = const()[name = tensor<string, []>("op_7406_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7406_end_0 = const()[name = tensor<string, []>("op_7406_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_7406_end_mask_0 = const()[name = tensor<string, []>("op_7406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7406_cast = slice_by_index(begin = var_7406_begin_0, end = var_7406_end_0, end_mask = var_7406_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7406_cast")];
            tensor<int32, [4]> var_7410_begin_0 = const()[name = tensor<string, []>("op_7410_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7410_end_0 = const()[name = tensor<string, []>("op_7410_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_7410_end_mask_0 = const()[name = tensor<string, []>("op_7410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7410_cast = slice_by_index(begin = var_7410_begin_0, end = var_7410_end_0, end_mask = var_7410_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7410_cast")];
            tensor<int32, [4]> var_7414_begin_0 = const()[name = tensor<string, []>("op_7414_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7414_end_0 = const()[name = tensor<string, []>("op_7414_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_7414_end_mask_0 = const()[name = tensor<string, []>("op_7414_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7414_cast = slice_by_index(begin = var_7414_begin_0, end = var_7414_end_0, end_mask = var_7414_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7414_cast")];
            tensor<int32, [4]> var_7418_begin_0 = const()[name = tensor<string, []>("op_7418_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7418_end_0 = const()[name = tensor<string, []>("op_7418_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7418_end_mask_0 = const()[name = tensor<string, []>("op_7418_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7418_cast = slice_by_index(begin = var_7418_begin_0, end = var_7418_end_0, end_mask = var_7418_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7418_cast")];
            tensor<int32, [4]> var_7422_begin_0 = const()[name = tensor<string, []>("op_7422_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7422_end_0 = const()[name = tensor<string, []>("op_7422_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7422_end_mask_0 = const()[name = tensor<string, []>("op_7422_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7422_cast = slice_by_index(begin = var_7422_begin_0, end = var_7422_end_0, end_mask = var_7422_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7422_cast")];
            tensor<int32, [4]> var_7426_begin_0 = const()[name = tensor<string, []>("op_7426_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7426_end_0 = const()[name = tensor<string, []>("op_7426_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_7426_end_mask_0 = const()[name = tensor<string, []>("op_7426_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7426_cast = slice_by_index(begin = var_7426_begin_0, end = var_7426_end_0, end_mask = var_7426_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7426_cast")];
            tensor<int32, [4]> var_7430_begin_0 = const()[name = tensor<string, []>("op_7430_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7430_end_0 = const()[name = tensor<string, []>("op_7430_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_7430_end_mask_0 = const()[name = tensor<string, []>("op_7430_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7430_cast = slice_by_index(begin = var_7430_begin_0, end = var_7430_end_0, end_mask = var_7430_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7430_cast")];
            tensor<int32, [4]> var_7434_begin_0 = const()[name = tensor<string, []>("op_7434_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7434_end_0 = const()[name = tensor<string, []>("op_7434_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_7434_end_mask_0 = const()[name = tensor<string, []>("op_7434_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7434_cast = slice_by_index(begin = var_7434_begin_0, end = var_7434_end_0, end_mask = var_7434_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7434_cast")];
            tensor<int32, [4]> var_7438_begin_0 = const()[name = tensor<string, []>("op_7438_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7438_end_0 = const()[name = tensor<string, []>("op_7438_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7438_end_mask_0 = const()[name = tensor<string, []>("op_7438_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7438_cast = slice_by_index(begin = var_7438_begin_0, end = var_7438_end_0, end_mask = var_7438_end_mask_0, x = q_29_cast)[name = tensor<string, []>("op_7438_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_7445_begin_0 = const()[name = tensor<string, []>("op_7445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7445_end_0 = const()[name = tensor<string, []>("op_7445_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_7445_end_mask_0 = const()[name = tensor<string, []>("op_7445_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, 64]> var_7445_cast = slice_by_index(begin = var_7445_begin_0, end = var_7445_end_0, end_mask = var_7445_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7445_cast")];
            tensor<int32, [4]> var_7449_begin_0 = const()[name = tensor<string, []>("op_7449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_7449_end_0 = const()[name = tensor<string, []>("op_7449_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_7449_end_mask_0 = const()[name = tensor<string, []>("op_7449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7449_cast = slice_by_index(begin = var_7449_begin_0, end = var_7449_end_0, end_mask = var_7449_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7449_cast")];
            tensor<int32, [4]> var_7453_begin_0 = const()[name = tensor<string, []>("op_7453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_7453_end_0 = const()[name = tensor<string, []>("op_7453_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_7453_end_mask_0 = const()[name = tensor<string, []>("op_7453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7453_cast = slice_by_index(begin = var_7453_begin_0, end = var_7453_end_0, end_mask = var_7453_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7453_cast")];
            tensor<int32, [4]> var_7457_begin_0 = const()[name = tensor<string, []>("op_7457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_7457_end_0 = const()[name = tensor<string, []>("op_7457_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7457_end_mask_0 = const()[name = tensor<string, []>("op_7457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7457_cast = slice_by_index(begin = var_7457_begin_0, end = var_7457_end_0, end_mask = var_7457_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7457_cast")];
            tensor<int32, [4]> var_7461_begin_0 = const()[name = tensor<string, []>("op_7461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_7461_end_0 = const()[name = tensor<string, []>("op_7461_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_7461_end_mask_0 = const()[name = tensor<string, []>("op_7461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7461_cast = slice_by_index(begin = var_7461_begin_0, end = var_7461_end_0, end_mask = var_7461_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7461_cast")];
            tensor<int32, [4]> var_7465_begin_0 = const()[name = tensor<string, []>("op_7465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7465_end_0 = const()[name = tensor<string, []>("op_7465_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_7465_end_mask_0 = const()[name = tensor<string, []>("op_7465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7465_cast = slice_by_index(begin = var_7465_begin_0, end = var_7465_end_0, end_mask = var_7465_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7465_cast")];
            tensor<int32, [4]> var_7469_begin_0 = const()[name = tensor<string, []>("op_7469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_7469_end_0 = const()[name = tensor<string, []>("op_7469_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_7469_end_mask_0 = const()[name = tensor<string, []>("op_7469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7469_cast = slice_by_index(begin = var_7469_begin_0, end = var_7469_end_0, end_mask = var_7469_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7469_cast")];
            tensor<int32, [4]> var_7473_begin_0 = const()[name = tensor<string, []>("op_7473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_7473_end_0 = const()[name = tensor<string, []>("op_7473_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_7473_end_mask_0 = const()[name = tensor<string, []>("op_7473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7473_cast = slice_by_index(begin = var_7473_begin_0, end = var_7473_end_0, end_mask = var_7473_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7473_cast")];
            tensor<int32, [4]> var_7477_begin_0 = const()[name = tensor<string, []>("op_7477_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_7477_end_0 = const()[name = tensor<string, []>("op_7477_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_7477_end_mask_0 = const()[name = tensor<string, []>("op_7477_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7477_cast = slice_by_index(begin = var_7477_begin_0, end = var_7477_end_0, end_mask = var_7477_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7477_cast")];
            tensor<int32, [4]> var_7481_begin_0 = const()[name = tensor<string, []>("op_7481_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_7481_end_0 = const()[name = tensor<string, []>("op_7481_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_7481_end_mask_0 = const()[name = tensor<string, []>("op_7481_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7481_cast = slice_by_index(begin = var_7481_begin_0, end = var_7481_end_0, end_mask = var_7481_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7481_cast")];
            tensor<int32, [4]> var_7485_begin_0 = const()[name = tensor<string, []>("op_7485_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7485_end_0 = const()[name = tensor<string, []>("op_7485_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_7485_end_mask_0 = const()[name = tensor<string, []>("op_7485_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7485_cast = slice_by_index(begin = var_7485_begin_0, end = var_7485_end_0, end_mask = var_7485_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7485_cast")];
            tensor<int32, [4]> var_7489_begin_0 = const()[name = tensor<string, []>("op_7489_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_7489_end_0 = const()[name = tensor<string, []>("op_7489_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_7489_end_mask_0 = const()[name = tensor<string, []>("op_7489_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7489_cast = slice_by_index(begin = var_7489_begin_0, end = var_7489_end_0, end_mask = var_7489_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7489_cast")];
            tensor<int32, [4]> var_7493_begin_0 = const()[name = tensor<string, []>("op_7493_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_7493_end_0 = const()[name = tensor<string, []>("op_7493_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_7493_end_mask_0 = const()[name = tensor<string, []>("op_7493_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7493_cast = slice_by_index(begin = var_7493_begin_0, end = var_7493_end_0, end_mask = var_7493_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7493_cast")];
            tensor<int32, [4]> var_7497_begin_0 = const()[name = tensor<string, []>("op_7497_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_7497_end_0 = const()[name = tensor<string, []>("op_7497_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_7497_end_mask_0 = const()[name = tensor<string, []>("op_7497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7497_cast = slice_by_index(begin = var_7497_begin_0, end = var_7497_end_0, end_mask = var_7497_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7497_cast")];
            tensor<int32, [4]> var_7501_begin_0 = const()[name = tensor<string, []>("op_7501_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_7501_end_0 = const()[name = tensor<string, []>("op_7501_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_7501_end_mask_0 = const()[name = tensor<string, []>("op_7501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7501_cast = slice_by_index(begin = var_7501_begin_0, end = var_7501_end_0, end_mask = var_7501_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7501_cast")];
            tensor<int32, [4]> var_7505_begin_0 = const()[name = tensor<string, []>("op_7505_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7505_end_0 = const()[name = tensor<string, []>("op_7505_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_7505_end_mask_0 = const()[name = tensor<string, []>("op_7505_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7505_cast = slice_by_index(begin = var_7505_begin_0, end = var_7505_end_0, end_mask = var_7505_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7505_cast")];
            tensor<int32, [4]> var_7509_begin_0 = const()[name = tensor<string, []>("op_7509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7509_end_0 = const()[name = tensor<string, []>("op_7509_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_7509_end_mask_0 = const()[name = tensor<string, []>("op_7509_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7509_cast = slice_by_index(begin = var_7509_begin_0, end = var_7509_end_0, end_mask = var_7509_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7509_cast")];
            tensor<int32, [4]> var_7513_begin_0 = const()[name = tensor<string, []>("op_7513_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7513_end_0 = const()[name = tensor<string, []>("op_7513_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_7513_end_mask_0 = const()[name = tensor<string, []>("op_7513_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7513_cast = slice_by_index(begin = var_7513_begin_0, end = var_7513_end_0, end_mask = var_7513_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7513_cast")];
            tensor<int32, [4]> var_7517_begin_0 = const()[name = tensor<string, []>("op_7517_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7517_end_0 = const()[name = tensor<string, []>("op_7517_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_7517_end_mask_0 = const()[name = tensor<string, []>("op_7517_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7517_cast = slice_by_index(begin = var_7517_begin_0, end = var_7517_end_0, end_mask = var_7517_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7517_cast")];
            tensor<int32, [4]> var_7521_begin_0 = const()[name = tensor<string, []>("op_7521_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7521_end_0 = const()[name = tensor<string, []>("op_7521_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_7521_end_mask_0 = const()[name = tensor<string, []>("op_7521_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_7521_cast = slice_by_index(begin = var_7521_begin_0, end = var_7521_end_0, end_mask = var_7521_end_mask_0, x = transpose_17)[name = tensor<string, []>("op_7521_cast")];
            tensor<int32, [4]> var_7523_begin_0 = const()[name = tensor<string, []>("op_7523_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7523_end_0 = const()[name = tensor<string, []>("op_7523_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_7523_end_mask_0 = const()[name = tensor<string, []>("op_7523_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7523_cast = slice_by_index(begin = var_7523_begin_0, end = var_7523_end_0, end_mask = var_7523_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7523_cast")];
            tensor<int32, [4]> var_7527_begin_0 = const()[name = tensor<string, []>("op_7527_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7527_end_0 = const()[name = tensor<string, []>("op_7527_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_7527_end_mask_0 = const()[name = tensor<string, []>("op_7527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7527_cast = slice_by_index(begin = var_7527_begin_0, end = var_7527_end_0, end_mask = var_7527_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7527_cast")];
            tensor<int32, [4]> var_7531_begin_0 = const()[name = tensor<string, []>("op_7531_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7531_end_0 = const()[name = tensor<string, []>("op_7531_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_7531_end_mask_0 = const()[name = tensor<string, []>("op_7531_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7531_cast = slice_by_index(begin = var_7531_begin_0, end = var_7531_end_0, end_mask = var_7531_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7531_cast")];
            tensor<int32, [4]> var_7535_begin_0 = const()[name = tensor<string, []>("op_7535_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7535_end_0 = const()[name = tensor<string, []>("op_7535_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7535_end_mask_0 = const()[name = tensor<string, []>("op_7535_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7535_cast = slice_by_index(begin = var_7535_begin_0, end = var_7535_end_0, end_mask = var_7535_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7535_cast")];
            tensor<int32, [4]> var_7539_begin_0 = const()[name = tensor<string, []>("op_7539_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7539_end_0 = const()[name = tensor<string, []>("op_7539_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7539_end_mask_0 = const()[name = tensor<string, []>("op_7539_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7539_cast = slice_by_index(begin = var_7539_begin_0, end = var_7539_end_0, end_mask = var_7539_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7539_cast")];
            tensor<int32, [4]> var_7543_begin_0 = const()[name = tensor<string, []>("op_7543_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7543_end_0 = const()[name = tensor<string, []>("op_7543_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_7543_end_mask_0 = const()[name = tensor<string, []>("op_7543_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7543_cast = slice_by_index(begin = var_7543_begin_0, end = var_7543_end_0, end_mask = var_7543_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7543_cast")];
            tensor<int32, [4]> var_7547_begin_0 = const()[name = tensor<string, []>("op_7547_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7547_end_0 = const()[name = tensor<string, []>("op_7547_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_7547_end_mask_0 = const()[name = tensor<string, []>("op_7547_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7547_cast = slice_by_index(begin = var_7547_begin_0, end = var_7547_end_0, end_mask = var_7547_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7547_cast")];
            tensor<int32, [4]> var_7551_begin_0 = const()[name = tensor<string, []>("op_7551_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7551_end_0 = const()[name = tensor<string, []>("op_7551_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_7551_end_mask_0 = const()[name = tensor<string, []>("op_7551_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7551_cast = slice_by_index(begin = var_7551_begin_0, end = var_7551_end_0, end_mask = var_7551_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7551_cast")];
            tensor<int32, [4]> var_7555_begin_0 = const()[name = tensor<string, []>("op_7555_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7555_end_0 = const()[name = tensor<string, []>("op_7555_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_7555_end_mask_0 = const()[name = tensor<string, []>("op_7555_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7555_cast = slice_by_index(begin = var_7555_begin_0, end = var_7555_end_0, end_mask = var_7555_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7555_cast")];
            tensor<int32, [4]> var_7559_begin_0 = const()[name = tensor<string, []>("op_7559_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7559_end_0 = const()[name = tensor<string, []>("op_7559_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7559_end_mask_0 = const()[name = tensor<string, []>("op_7559_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7559_cast = slice_by_index(begin = var_7559_begin_0, end = var_7559_end_0, end_mask = var_7559_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7559_cast")];
            tensor<int32, [4]> var_7563_begin_0 = const()[name = tensor<string, []>("op_7563_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7563_end_0 = const()[name = tensor<string, []>("op_7563_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_7563_end_mask_0 = const()[name = tensor<string, []>("op_7563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7563_cast = slice_by_index(begin = var_7563_begin_0, end = var_7563_end_0, end_mask = var_7563_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7563_cast")];
            tensor<int32, [4]> var_7567_begin_0 = const()[name = tensor<string, []>("op_7567_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7567_end_0 = const()[name = tensor<string, []>("op_7567_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_7567_end_mask_0 = const()[name = tensor<string, []>("op_7567_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7567_cast = slice_by_index(begin = var_7567_begin_0, end = var_7567_end_0, end_mask = var_7567_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7567_cast")];
            tensor<int32, [4]> var_7571_begin_0 = const()[name = tensor<string, []>("op_7571_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7571_end_0 = const()[name = tensor<string, []>("op_7571_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_7571_end_mask_0 = const()[name = tensor<string, []>("op_7571_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7571_cast = slice_by_index(begin = var_7571_begin_0, end = var_7571_end_0, end_mask = var_7571_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7571_cast")];
            tensor<int32, [4]> var_7575_begin_0 = const()[name = tensor<string, []>("op_7575_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7575_end_0 = const()[name = tensor<string, []>("op_7575_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_7575_end_mask_0 = const()[name = tensor<string, []>("op_7575_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7575_cast = slice_by_index(begin = var_7575_begin_0, end = var_7575_end_0, end_mask = var_7575_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7575_cast")];
            tensor<int32, [4]> var_7579_begin_0 = const()[name = tensor<string, []>("op_7579_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7579_end_0 = const()[name = tensor<string, []>("op_7579_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7579_end_mask_0 = const()[name = tensor<string, []>("op_7579_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7579_cast = slice_by_index(begin = var_7579_begin_0, end = var_7579_end_0, end_mask = var_7579_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7579_cast")];
            tensor<int32, [4]> var_7583_begin_0 = const()[name = tensor<string, []>("op_7583_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7583_end_0 = const()[name = tensor<string, []>("op_7583_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7583_end_mask_0 = const()[name = tensor<string, []>("op_7583_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7583_cast = slice_by_index(begin = var_7583_begin_0, end = var_7583_end_0, end_mask = var_7583_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7583_cast")];
            tensor<int32, [4]> var_7587_begin_0 = const()[name = tensor<string, []>("op_7587_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7587_end_0 = const()[name = tensor<string, []>("op_7587_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_7587_end_mask_0 = const()[name = tensor<string, []>("op_7587_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7587_cast = slice_by_index(begin = var_7587_begin_0, end = var_7587_end_0, end_mask = var_7587_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7587_cast")];
            tensor<int32, [4]> var_7591_begin_0 = const()[name = tensor<string, []>("op_7591_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7591_end_0 = const()[name = tensor<string, []>("op_7591_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_7591_end_mask_0 = const()[name = tensor<string, []>("op_7591_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7591_cast = slice_by_index(begin = var_7591_begin_0, end = var_7591_end_0, end_mask = var_7591_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7591_cast")];
            tensor<int32, [4]> var_7595_begin_0 = const()[name = tensor<string, []>("op_7595_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7595_end_0 = const()[name = tensor<string, []>("op_7595_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_7595_end_mask_0 = const()[name = tensor<string, []>("op_7595_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7595_cast = slice_by_index(begin = var_7595_begin_0, end = var_7595_end_0, end_mask = var_7595_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7595_cast")];
            tensor<int32, [4]> var_7599_begin_0 = const()[name = tensor<string, []>("op_7599_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7599_end_0 = const()[name = tensor<string, []>("op_7599_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7599_end_mask_0 = const()[name = tensor<string, []>("op_7599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7599_cast = slice_by_index(begin = var_7599_begin_0, end = var_7599_end_0, end_mask = var_7599_end_mask_0, x = v_29_cast)[name = tensor<string, []>("op_7599_cast")];
            tensor<string, []> var_7603_equation_0 = const()[name = tensor<string, []>("op_7603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7603_cast = einsum(equation = var_7603_equation_0, values = (var_7445_cast, var_7362_cast))[name = tensor<string, []>("op_7603_cast")];
            tensor<fp16, []> var_7604_to_fp16 = const()[name = tensor<string, []>("op_7604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_241_cast = mul(x = var_7603_cast, y = var_7604_to_fp16)[name = tensor<string, []>("aw_241_cast")];
            tensor<string, []> var_7607_equation_0 = const()[name = tensor<string, []>("op_7607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7607_cast = einsum(equation = var_7607_equation_0, values = (var_7449_cast, var_7366_cast))[name = tensor<string, []>("op_7607_cast")];
            tensor<fp16, []> var_7608_to_fp16 = const()[name = tensor<string, []>("op_7608_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_243_cast = mul(x = var_7607_cast, y = var_7608_to_fp16)[name = tensor<string, []>("aw_243_cast")];
            tensor<string, []> var_7611_equation_0 = const()[name = tensor<string, []>("op_7611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7611_cast = einsum(equation = var_7611_equation_0, values = (var_7453_cast, var_7370_cast))[name = tensor<string, []>("op_7611_cast")];
            tensor<fp16, []> var_7612_to_fp16 = const()[name = tensor<string, []>("op_7612_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_245_cast = mul(x = var_7611_cast, y = var_7612_to_fp16)[name = tensor<string, []>("aw_245_cast")];
            tensor<string, []> var_7615_equation_0 = const()[name = tensor<string, []>("op_7615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7615_cast = einsum(equation = var_7615_equation_0, values = (var_7457_cast, var_7374_cast))[name = tensor<string, []>("op_7615_cast")];
            tensor<fp16, []> var_7616_to_fp16 = const()[name = tensor<string, []>("op_7616_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_247_cast = mul(x = var_7615_cast, y = var_7616_to_fp16)[name = tensor<string, []>("aw_247_cast")];
            tensor<string, []> var_7619_equation_0 = const()[name = tensor<string, []>("op_7619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7619_cast = einsum(equation = var_7619_equation_0, values = (var_7461_cast, var_7378_cast))[name = tensor<string, []>("op_7619_cast")];
            tensor<fp16, []> var_7620_to_fp16 = const()[name = tensor<string, []>("op_7620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_249_cast = mul(x = var_7619_cast, y = var_7620_to_fp16)[name = tensor<string, []>("aw_249_cast")];
            tensor<string, []> var_7623_equation_0 = const()[name = tensor<string, []>("op_7623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7623_cast = einsum(equation = var_7623_equation_0, values = (var_7465_cast, var_7382_cast))[name = tensor<string, []>("op_7623_cast")];
            tensor<fp16, []> var_7624_to_fp16 = const()[name = tensor<string, []>("op_7624_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_251_cast = mul(x = var_7623_cast, y = var_7624_to_fp16)[name = tensor<string, []>("aw_251_cast")];
            tensor<string, []> var_7627_equation_0 = const()[name = tensor<string, []>("op_7627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7627_cast = einsum(equation = var_7627_equation_0, values = (var_7469_cast, var_7386_cast))[name = tensor<string, []>("op_7627_cast")];
            tensor<fp16, []> var_7628_to_fp16 = const()[name = tensor<string, []>("op_7628_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_253_cast = mul(x = var_7627_cast, y = var_7628_to_fp16)[name = tensor<string, []>("aw_253_cast")];
            tensor<string, []> var_7631_equation_0 = const()[name = tensor<string, []>("op_7631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7631_cast = einsum(equation = var_7631_equation_0, values = (var_7473_cast, var_7390_cast))[name = tensor<string, []>("op_7631_cast")];
            tensor<fp16, []> var_7632_to_fp16 = const()[name = tensor<string, []>("op_7632_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_255_cast = mul(x = var_7631_cast, y = var_7632_to_fp16)[name = tensor<string, []>("aw_255_cast")];
            tensor<string, []> var_7635_equation_0 = const()[name = tensor<string, []>("op_7635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7635_cast = einsum(equation = var_7635_equation_0, values = (var_7477_cast, var_7394_cast))[name = tensor<string, []>("op_7635_cast")];
            tensor<fp16, []> var_7636_to_fp16 = const()[name = tensor<string, []>("op_7636_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_257_cast = mul(x = var_7635_cast, y = var_7636_to_fp16)[name = tensor<string, []>("aw_257_cast")];
            tensor<string, []> var_7639_equation_0 = const()[name = tensor<string, []>("op_7639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7639_cast = einsum(equation = var_7639_equation_0, values = (var_7481_cast, var_7398_cast))[name = tensor<string, []>("op_7639_cast")];
            tensor<fp16, []> var_7640_to_fp16 = const()[name = tensor<string, []>("op_7640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_259_cast = mul(x = var_7639_cast, y = var_7640_to_fp16)[name = tensor<string, []>("aw_259_cast")];
            tensor<string, []> var_7643_equation_0 = const()[name = tensor<string, []>("op_7643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7643_cast = einsum(equation = var_7643_equation_0, values = (var_7485_cast, var_7402_cast))[name = tensor<string, []>("op_7643_cast")];
            tensor<fp16, []> var_7644_to_fp16 = const()[name = tensor<string, []>("op_7644_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_261_cast = mul(x = var_7643_cast, y = var_7644_to_fp16)[name = tensor<string, []>("aw_261_cast")];
            tensor<string, []> var_7647_equation_0 = const()[name = tensor<string, []>("op_7647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7647_cast = einsum(equation = var_7647_equation_0, values = (var_7489_cast, var_7406_cast))[name = tensor<string, []>("op_7647_cast")];
            tensor<fp16, []> var_7648_to_fp16 = const()[name = tensor<string, []>("op_7648_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_263_cast = mul(x = var_7647_cast, y = var_7648_to_fp16)[name = tensor<string, []>("aw_263_cast")];
            tensor<string, []> var_7651_equation_0 = const()[name = tensor<string, []>("op_7651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7651_cast = einsum(equation = var_7651_equation_0, values = (var_7493_cast, var_7410_cast))[name = tensor<string, []>("op_7651_cast")];
            tensor<fp16, []> var_7652_to_fp16 = const()[name = tensor<string, []>("op_7652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_265_cast = mul(x = var_7651_cast, y = var_7652_to_fp16)[name = tensor<string, []>("aw_265_cast")];
            tensor<string, []> var_7655_equation_0 = const()[name = tensor<string, []>("op_7655_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7655_cast = einsum(equation = var_7655_equation_0, values = (var_7497_cast, var_7414_cast))[name = tensor<string, []>("op_7655_cast")];
            tensor<fp16, []> var_7656_to_fp16 = const()[name = tensor<string, []>("op_7656_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_267_cast = mul(x = var_7655_cast, y = var_7656_to_fp16)[name = tensor<string, []>("aw_267_cast")];
            tensor<string, []> var_7659_equation_0 = const()[name = tensor<string, []>("op_7659_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7659_cast = einsum(equation = var_7659_equation_0, values = (var_7501_cast, var_7418_cast))[name = tensor<string, []>("op_7659_cast")];
            tensor<fp16, []> var_7660_to_fp16 = const()[name = tensor<string, []>("op_7660_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_269_cast = mul(x = var_7659_cast, y = var_7660_to_fp16)[name = tensor<string, []>("aw_269_cast")];
            tensor<string, []> var_7663_equation_0 = const()[name = tensor<string, []>("op_7663_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7663_cast = einsum(equation = var_7663_equation_0, values = (var_7505_cast, var_7422_cast))[name = tensor<string, []>("op_7663_cast")];
            tensor<fp16, []> var_7664_to_fp16 = const()[name = tensor<string, []>("op_7664_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_271_cast = mul(x = var_7663_cast, y = var_7664_to_fp16)[name = tensor<string, []>("aw_271_cast")];
            tensor<string, []> var_7667_equation_0 = const()[name = tensor<string, []>("op_7667_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7667_cast = einsum(equation = var_7667_equation_0, values = (var_7509_cast, var_7426_cast))[name = tensor<string, []>("op_7667_cast")];
            tensor<fp16, []> var_7668_to_fp16 = const()[name = tensor<string, []>("op_7668_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_273_cast = mul(x = var_7667_cast, y = var_7668_to_fp16)[name = tensor<string, []>("aw_273_cast")];
            tensor<string, []> var_7671_equation_0 = const()[name = tensor<string, []>("op_7671_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7671_cast = einsum(equation = var_7671_equation_0, values = (var_7513_cast, var_7430_cast))[name = tensor<string, []>("op_7671_cast")];
            tensor<fp16, []> var_7672_to_fp16 = const()[name = tensor<string, []>("op_7672_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_275_cast = mul(x = var_7671_cast, y = var_7672_to_fp16)[name = tensor<string, []>("aw_275_cast")];
            tensor<string, []> var_7675_equation_0 = const()[name = tensor<string, []>("op_7675_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7675_cast = einsum(equation = var_7675_equation_0, values = (var_7517_cast, var_7434_cast))[name = tensor<string, []>("op_7675_cast")];
            tensor<fp16, []> var_7676_to_fp16 = const()[name = tensor<string, []>("op_7676_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_277_cast = mul(x = var_7675_cast, y = var_7676_to_fp16)[name = tensor<string, []>("aw_277_cast")];
            tensor<string, []> var_7679_equation_0 = const()[name = tensor<string, []>("op_7679_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_7679_cast = einsum(equation = var_7679_equation_0, values = (var_7521_cast, var_7438_cast))[name = tensor<string, []>("op_7679_cast")];
            tensor<fp16, []> var_7680_to_fp16 = const()[name = tensor<string, []>("op_7680_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_279_cast = mul(x = var_7679_cast, y = var_7680_to_fp16)[name = tensor<string, []>("aw_279_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7682_cast = softmax(axis = var_7228, x = aw_241_cast)[name = tensor<string, []>("op_7682_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7683_cast = softmax(axis = var_7228, x = aw_243_cast)[name = tensor<string, []>("op_7683_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7684_cast = softmax(axis = var_7228, x = aw_245_cast)[name = tensor<string, []>("op_7684_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7685_cast = softmax(axis = var_7228, x = aw_247_cast)[name = tensor<string, []>("op_7685_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7686_cast = softmax(axis = var_7228, x = aw_249_cast)[name = tensor<string, []>("op_7686_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7687_cast = softmax(axis = var_7228, x = aw_251_cast)[name = tensor<string, []>("op_7687_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7688_cast = softmax(axis = var_7228, x = aw_253_cast)[name = tensor<string, []>("op_7688_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7689_cast = softmax(axis = var_7228, x = aw_255_cast)[name = tensor<string, []>("op_7689_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7690_cast = softmax(axis = var_7228, x = aw_257_cast)[name = tensor<string, []>("op_7690_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7691_cast = softmax(axis = var_7228, x = aw_259_cast)[name = tensor<string, []>("op_7691_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7692_cast = softmax(axis = var_7228, x = aw_261_cast)[name = tensor<string, []>("op_7692_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7693_cast = softmax(axis = var_7228, x = aw_263_cast)[name = tensor<string, []>("op_7693_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7694_cast = softmax(axis = var_7228, x = aw_265_cast)[name = tensor<string, []>("op_7694_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7695_cast = softmax(axis = var_7228, x = aw_267_cast)[name = tensor<string, []>("op_7695_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7696_cast = softmax(axis = var_7228, x = aw_269_cast)[name = tensor<string, []>("op_7696_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7697_cast = softmax(axis = var_7228, x = aw_271_cast)[name = tensor<string, []>("op_7697_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7698_cast = softmax(axis = var_7228, x = aw_273_cast)[name = tensor<string, []>("op_7698_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7699_cast = softmax(axis = var_7228, x = aw_275_cast)[name = tensor<string, []>("op_7699_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7700_cast = softmax(axis = var_7228, x = aw_277_cast)[name = tensor<string, []>("op_7700_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_7701_cast = softmax(axis = var_7228, x = aw_279_cast)[name = tensor<string, []>("op_7701_cast")];
            tensor<string, []> var_7703_equation_0 = const()[name = tensor<string, []>("op_7703_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7703_cast = einsum(equation = var_7703_equation_0, values = (var_7523_cast, var_7682_cast))[name = tensor<string, []>("op_7703_cast")];
            tensor<string, []> var_7705_equation_0 = const()[name = tensor<string, []>("op_7705_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7705_cast = einsum(equation = var_7705_equation_0, values = (var_7527_cast, var_7683_cast))[name = tensor<string, []>("op_7705_cast")];
            tensor<string, []> var_7707_equation_0 = const()[name = tensor<string, []>("op_7707_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7707_cast = einsum(equation = var_7707_equation_0, values = (var_7531_cast, var_7684_cast))[name = tensor<string, []>("op_7707_cast")];
            tensor<string, []> var_7709_equation_0 = const()[name = tensor<string, []>("op_7709_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7709_cast = einsum(equation = var_7709_equation_0, values = (var_7535_cast, var_7685_cast))[name = tensor<string, []>("op_7709_cast")];
            tensor<string, []> var_7711_equation_0 = const()[name = tensor<string, []>("op_7711_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7711_cast = einsum(equation = var_7711_equation_0, values = (var_7539_cast, var_7686_cast))[name = tensor<string, []>("op_7711_cast")];
            tensor<string, []> var_7713_equation_0 = const()[name = tensor<string, []>("op_7713_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7713_cast = einsum(equation = var_7713_equation_0, values = (var_7543_cast, var_7687_cast))[name = tensor<string, []>("op_7713_cast")];
            tensor<string, []> var_7715_equation_0 = const()[name = tensor<string, []>("op_7715_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7715_cast = einsum(equation = var_7715_equation_0, values = (var_7547_cast, var_7688_cast))[name = tensor<string, []>("op_7715_cast")];
            tensor<string, []> var_7717_equation_0 = const()[name = tensor<string, []>("op_7717_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7717_cast = einsum(equation = var_7717_equation_0, values = (var_7551_cast, var_7689_cast))[name = tensor<string, []>("op_7717_cast")];
            tensor<string, []> var_7719_equation_0 = const()[name = tensor<string, []>("op_7719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7719_cast = einsum(equation = var_7719_equation_0, values = (var_7555_cast, var_7690_cast))[name = tensor<string, []>("op_7719_cast")];
            tensor<string, []> var_7721_equation_0 = const()[name = tensor<string, []>("op_7721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7721_cast = einsum(equation = var_7721_equation_0, values = (var_7559_cast, var_7691_cast))[name = tensor<string, []>("op_7721_cast")];
            tensor<string, []> var_7723_equation_0 = const()[name = tensor<string, []>("op_7723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7723_cast = einsum(equation = var_7723_equation_0, values = (var_7563_cast, var_7692_cast))[name = tensor<string, []>("op_7723_cast")];
            tensor<string, []> var_7725_equation_0 = const()[name = tensor<string, []>("op_7725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7725_cast = einsum(equation = var_7725_equation_0, values = (var_7567_cast, var_7693_cast))[name = tensor<string, []>("op_7725_cast")];
            tensor<string, []> var_7727_equation_0 = const()[name = tensor<string, []>("op_7727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7727_cast = einsum(equation = var_7727_equation_0, values = (var_7571_cast, var_7694_cast))[name = tensor<string, []>("op_7727_cast")];
            tensor<string, []> var_7729_equation_0 = const()[name = tensor<string, []>("op_7729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7729_cast = einsum(equation = var_7729_equation_0, values = (var_7575_cast, var_7695_cast))[name = tensor<string, []>("op_7729_cast")];
            tensor<string, []> var_7731_equation_0 = const()[name = tensor<string, []>("op_7731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7731_cast = einsum(equation = var_7731_equation_0, values = (var_7579_cast, var_7696_cast))[name = tensor<string, []>("op_7731_cast")];
            tensor<string, []> var_7733_equation_0 = const()[name = tensor<string, []>("op_7733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7733_cast = einsum(equation = var_7733_equation_0, values = (var_7583_cast, var_7697_cast))[name = tensor<string, []>("op_7733_cast")];
            tensor<string, []> var_7735_equation_0 = const()[name = tensor<string, []>("op_7735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7735_cast = einsum(equation = var_7735_equation_0, values = (var_7587_cast, var_7698_cast))[name = tensor<string, []>("op_7735_cast")];
            tensor<string, []> var_7737_equation_0 = const()[name = tensor<string, []>("op_7737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7737_cast = einsum(equation = var_7737_equation_0, values = (var_7591_cast, var_7699_cast))[name = tensor<string, []>("op_7737_cast")];
            tensor<string, []> var_7739_equation_0 = const()[name = tensor<string, []>("op_7739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7739_cast = einsum(equation = var_7739_equation_0, values = (var_7595_cast, var_7700_cast))[name = tensor<string, []>("op_7739_cast")];
            tensor<string, []> var_7741_equation_0 = const()[name = tensor<string, []>("op_7741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_7741_cast = einsum(equation = var_7741_equation_0, values = (var_7599_cast, var_7701_cast))[name = tensor<string, []>("op_7741_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_7228, interleave = input_301_interleave_0, values = (var_7703_cast, var_7705_cast, var_7707_cast, var_7709_cast, var_7711_cast, var_7713_cast, var_7715_cast, var_7717_cast, var_7719_cast, var_7721_cast, var_7723_cast, var_7725_cast, var_7727_cast, var_7729_cast, var_7731_cast, var_7733_cast, var_7735_cast, var_7737_cast, var_7739_cast, var_7741_cast))[name = tensor<string, []>("input_301_cast")];
            tensor<int32, [2]> var_7747 = const()[name = tensor<string, []>("op_7747"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7749 = const()[name = tensor<string, []>("op_7749"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_7751_pad_type_0 = const()[name = tensor<string, []>("op_7751_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_7751_pad_0 = const()[name = tensor<string, []>("op_7751_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142765888))), lut = tensor<fp16, [4]>([-0x1.684p-4, -0x1.aep-6, 0x1.b04p-6, 0x1.68p-4]), 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, []>(143175552)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7751_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_7749, groups = var_7228, pad = var_7751_pad_0, pad_type = var_7751_pad_type_0, strides = var_7747, 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_7751_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_45_cast = add(x = var_7751_cast, y = inputs_43_cast)[name = tensor<string, []>("inputs_45_cast")];
            tensor<int32, [1]> var_7755 = const()[name = tensor<string, []>("op_7755"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_45_cast = reduce_mean(axes = var_7755, keep_dims = var_7223, 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_7759 = const()[name = tensor<string, []>("op_7759"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_7760_cast = reduce_mean(axes = var_7759, keep_dims = var_7223, x = zero_mean_sq_45_cast)[name = tensor<string, []>("op_7760_cast")];
            tensor<fp16, []> var_7761_to_fp16 = const()[name = tensor<string, []>("op_7761_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_7762_cast = add(x = var_7760_cast, y = var_7761_to_fp16)[name = tensor<string, []>("op_7762_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_7762_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_7766_to_fp16 = const()[name = tensor<string, []>("op_7766_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143178176)))];
            tensor<fp16, [2, 1280, 1, 256]> var_7767_cast = add(x = out_45_cast, y = var_7766_to_fp16)[name = tensor<string, []>("op_7767_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_7769_to_fp16 = const()[name = tensor<string, []>("op_7769_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143180800)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_169_cast = mul(x = var_7767_cast, y = var_7769_to_fp16)[name = tensor<string, []>("hidden_states_169_cast")];
            tensor<int32, [2]> var_7776 = const()[name = tensor<string, []>("op_7776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7778 = const()[name = tensor<string, []>("op_7778"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143183424))), lut = tensor<fp16, [4]>([-0x1.10cp-4, -0x1.468p-6, 0x1.458p-6, 0x1.10cp-4]), 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_7778, groups = var_7228, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = var_7776, 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_7782 = const()[name = tensor<string, []>("op_7782"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7784 = const()[name = tensor<string, []>("op_7784"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_61_pad_type_0 = const()[name = tensor<string, []>("k_61_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_61_pad_0 = const()[name = tensor<string, []>("k_61_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143593088))), lut = tensor<fp16, [4]>([-0x1.1f8p-4, -0x1.55cp-6, 0x1.544p-6, 0x1.1f4p-4]), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_61_cast = conv(dilations = var_7784, groups = var_7228, pad = k_61_pad_0, pad_type = k_61_pad_type_0, strides = var_7782, 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_7788 = const()[name = tensor<string, []>("op_7788"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_7790 = const()[name = tensor<string, []>("op_7790"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_31_pad_type_0 = const()[name = tensor<string, []>("v_31_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_31_pad_0 = const()[name = tensor<string, []>("v_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143920832))), lut = tensor<fp16, [4]>([-0x1.bacp-5, -0x1.02p-6, 0x1.03p-6, 0x1.bc4p-5]), name = tensor<string, []>("up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_31_cast = conv(dilations = var_7790, groups = var_7228, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = var_7788, 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_7794_begin_0 = const()[name = tensor<string, []>("op_7794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7794_end_0 = const()[name = tensor<string, []>("op_7794_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_7794_end_mask_0 = const()[name = tensor<string, []>("op_7794_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7794_cast = slice_by_index(begin = var_7794_begin_0, end = var_7794_end_0, end_mask = var_7794_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7794_cast")];
            tensor<int32, [4]> var_7798_begin_0 = const()[name = tensor<string, []>("op_7798_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7798_end_0 = const()[name = tensor<string, []>("op_7798_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_7798_end_mask_0 = const()[name = tensor<string, []>("op_7798_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7798_cast = slice_by_index(begin = var_7798_begin_0, end = var_7798_end_0, end_mask = var_7798_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7798_cast")];
            tensor<int32, [4]> var_7802_begin_0 = const()[name = tensor<string, []>("op_7802_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7802_end_0 = const()[name = tensor<string, []>("op_7802_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_7802_end_mask_0 = const()[name = tensor<string, []>("op_7802_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7802_cast = slice_by_index(begin = var_7802_begin_0, end = var_7802_end_0, end_mask = var_7802_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7802_cast")];
            tensor<int32, [4]> var_7806_begin_0 = const()[name = tensor<string, []>("op_7806_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7806_end_0 = const()[name = tensor<string, []>("op_7806_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_7806_end_mask_0 = const()[name = tensor<string, []>("op_7806_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7806_cast = slice_by_index(begin = var_7806_begin_0, end = var_7806_end_0, end_mask = var_7806_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7806_cast")];
            tensor<int32, [4]> var_7810_begin_0 = const()[name = tensor<string, []>("op_7810_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7810_end_0 = const()[name = tensor<string, []>("op_7810_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_7810_end_mask_0 = const()[name = tensor<string, []>("op_7810_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7810_cast = slice_by_index(begin = var_7810_begin_0, end = var_7810_end_0, end_mask = var_7810_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7810_cast")];
            tensor<int32, [4]> var_7814_begin_0 = const()[name = tensor<string, []>("op_7814_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7814_end_0 = const()[name = tensor<string, []>("op_7814_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_7814_end_mask_0 = const()[name = tensor<string, []>("op_7814_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7814_cast = slice_by_index(begin = var_7814_begin_0, end = var_7814_end_0, end_mask = var_7814_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7814_cast")];
            tensor<int32, [4]> var_7818_begin_0 = const()[name = tensor<string, []>("op_7818_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7818_end_0 = const()[name = tensor<string, []>("op_7818_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_7818_end_mask_0 = const()[name = tensor<string, []>("op_7818_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7818_cast = slice_by_index(begin = var_7818_begin_0, end = var_7818_end_0, end_mask = var_7818_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7818_cast")];
            tensor<int32, [4]> var_7822_begin_0 = const()[name = tensor<string, []>("op_7822_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7822_end_0 = const()[name = tensor<string, []>("op_7822_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_7822_end_mask_0 = const()[name = tensor<string, []>("op_7822_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7822_cast = slice_by_index(begin = var_7822_begin_0, end = var_7822_end_0, end_mask = var_7822_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7822_cast")];
            tensor<int32, [4]> var_7826_begin_0 = const()[name = tensor<string, []>("op_7826_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7826_end_0 = const()[name = tensor<string, []>("op_7826_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_7826_end_mask_0 = const()[name = tensor<string, []>("op_7826_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7826_cast = slice_by_index(begin = var_7826_begin_0, end = var_7826_end_0, end_mask = var_7826_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7826_cast")];
            tensor<int32, [4]> var_7830_begin_0 = const()[name = tensor<string, []>("op_7830_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7830_end_0 = const()[name = tensor<string, []>("op_7830_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_7830_end_mask_0 = const()[name = tensor<string, []>("op_7830_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7830_cast = slice_by_index(begin = var_7830_begin_0, end = var_7830_end_0, end_mask = var_7830_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7830_cast")];
            tensor<int32, [4]> var_7834_begin_0 = const()[name = tensor<string, []>("op_7834_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7834_end_0 = const()[name = tensor<string, []>("op_7834_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_7834_end_mask_0 = const()[name = tensor<string, []>("op_7834_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7834_cast = slice_by_index(begin = var_7834_begin_0, end = var_7834_end_0, end_mask = var_7834_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7834_cast")];
            tensor<int32, [4]> var_7838_begin_0 = const()[name = tensor<string, []>("op_7838_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7838_end_0 = const()[name = tensor<string, []>("op_7838_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_7838_end_mask_0 = const()[name = tensor<string, []>("op_7838_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7838_cast = slice_by_index(begin = var_7838_begin_0, end = var_7838_end_0, end_mask = var_7838_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7838_cast")];
            tensor<int32, [4]> var_7842_begin_0 = const()[name = tensor<string, []>("op_7842_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_7842_end_0 = const()[name = tensor<string, []>("op_7842_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_7842_end_mask_0 = const()[name = tensor<string, []>("op_7842_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7842_cast = slice_by_index(begin = var_7842_begin_0, end = var_7842_end_0, end_mask = var_7842_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7842_cast")];
            tensor<int32, [4]> var_7846_begin_0 = const()[name = tensor<string, []>("op_7846_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_7846_end_0 = const()[name = tensor<string, []>("op_7846_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_7846_end_mask_0 = const()[name = tensor<string, []>("op_7846_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7846_cast = slice_by_index(begin = var_7846_begin_0, end = var_7846_end_0, end_mask = var_7846_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7846_cast")];
            tensor<int32, [4]> var_7850_begin_0 = const()[name = tensor<string, []>("op_7850_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_7850_end_0 = const()[name = tensor<string, []>("op_7850_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_7850_end_mask_0 = const()[name = tensor<string, []>("op_7850_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7850_cast = slice_by_index(begin = var_7850_begin_0, end = var_7850_end_0, end_mask = var_7850_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7850_cast")];
            tensor<int32, [4]> var_7854_begin_0 = const()[name = tensor<string, []>("op_7854_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_7854_end_0 = const()[name = tensor<string, []>("op_7854_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_7854_end_mask_0 = const()[name = tensor<string, []>("op_7854_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7854_cast = slice_by_index(begin = var_7854_begin_0, end = var_7854_end_0, end_mask = var_7854_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7854_cast")];
            tensor<int32, [4]> var_7858_begin_0 = const()[name = tensor<string, []>("op_7858_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_7858_end_0 = const()[name = tensor<string, []>("op_7858_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_7858_end_mask_0 = const()[name = tensor<string, []>("op_7858_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7858_cast = slice_by_index(begin = var_7858_begin_0, end = var_7858_end_0, end_mask = var_7858_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7858_cast")];
            tensor<int32, [4]> var_7862_begin_0 = const()[name = tensor<string, []>("op_7862_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_7862_end_0 = const()[name = tensor<string, []>("op_7862_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_7862_end_mask_0 = const()[name = tensor<string, []>("op_7862_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7862_cast = slice_by_index(begin = var_7862_begin_0, end = var_7862_end_0, end_mask = var_7862_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7862_cast")];
            tensor<int32, [4]> var_7866_begin_0 = const()[name = tensor<string, []>("op_7866_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_7866_end_0 = const()[name = tensor<string, []>("op_7866_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_7866_end_mask_0 = const()[name = tensor<string, []>("op_7866_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7866_cast = slice_by_index(begin = var_7866_begin_0, end = var_7866_end_0, end_mask = var_7866_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7866_cast")];
            tensor<int32, [4]> var_7870_begin_0 = const()[name = tensor<string, []>("op_7870_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_7870_end_0 = const()[name = tensor<string, []>("op_7870_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_7870_end_mask_0 = const()[name = tensor<string, []>("op_7870_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_7870_cast = slice_by_index(begin = var_7870_begin_0, end = var_7870_end_0, end_mask = var_7870_end_mask_0, x = q_31_cast)[name = tensor<string, []>("op_7870_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_7877_begin_0 = const()[name = tensor<string, []>("op_7877_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7877_end_0 = const()[name = tensor<string, []>("op_7877_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_7877_end_mask_0 = const()[name = tensor<string, []>("op_7877_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, 64]> var_7877_cast = slice_by_index(begin = var_7877_begin_0, end = var_7877_end_0, end_mask = var_7877_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7877_cast")];
            tensor<int32, [4]> var_7881_begin_0 = const()[name = tensor<string, []>("op_7881_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_7881_end_0 = const()[name = tensor<string, []>("op_7881_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_7881_end_mask_0 = const()[name = tensor<string, []>("op_7881_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7881_cast = slice_by_index(begin = var_7881_begin_0, end = var_7881_end_0, end_mask = var_7881_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7881_cast")];
            tensor<int32, [4]> var_7885_begin_0 = const()[name = tensor<string, []>("op_7885_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_7885_end_0 = const()[name = tensor<string, []>("op_7885_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_7885_end_mask_0 = const()[name = tensor<string, []>("op_7885_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7885_cast = slice_by_index(begin = var_7885_begin_0, end = var_7885_end_0, end_mask = var_7885_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7885_cast")];
            tensor<int32, [4]> var_7889_begin_0 = const()[name = tensor<string, []>("op_7889_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_7889_end_0 = const()[name = tensor<string, []>("op_7889_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_7889_end_mask_0 = const()[name = tensor<string, []>("op_7889_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7889_cast = slice_by_index(begin = var_7889_begin_0, end = var_7889_end_0, end_mask = var_7889_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7889_cast")];
            tensor<int32, [4]> var_7893_begin_0 = const()[name = tensor<string, []>("op_7893_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_7893_end_0 = const()[name = tensor<string, []>("op_7893_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_7893_end_mask_0 = const()[name = tensor<string, []>("op_7893_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7893_cast = slice_by_index(begin = var_7893_begin_0, end = var_7893_end_0, end_mask = var_7893_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7893_cast")];
            tensor<int32, [4]> var_7897_begin_0 = const()[name = tensor<string, []>("op_7897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_7897_end_0 = const()[name = tensor<string, []>("op_7897_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_7897_end_mask_0 = const()[name = tensor<string, []>("op_7897_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7897_cast = slice_by_index(begin = var_7897_begin_0, end = var_7897_end_0, end_mask = var_7897_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7897_cast")];
            tensor<int32, [4]> var_7901_begin_0 = const()[name = tensor<string, []>("op_7901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_7901_end_0 = const()[name = tensor<string, []>("op_7901_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_7901_end_mask_0 = const()[name = tensor<string, []>("op_7901_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7901_cast = slice_by_index(begin = var_7901_begin_0, end = var_7901_end_0, end_mask = var_7901_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7901_cast")];
            tensor<int32, [4]> var_7905_begin_0 = const()[name = tensor<string, []>("op_7905_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_7905_end_0 = const()[name = tensor<string, []>("op_7905_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_7905_end_mask_0 = const()[name = tensor<string, []>("op_7905_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7905_cast = slice_by_index(begin = var_7905_begin_0, end = var_7905_end_0, end_mask = var_7905_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7905_cast")];
            tensor<int32, [4]> var_7909_begin_0 = const()[name = tensor<string, []>("op_7909_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_7909_end_0 = const()[name = tensor<string, []>("op_7909_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_7909_end_mask_0 = const()[name = tensor<string, []>("op_7909_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7909_cast = slice_by_index(begin = var_7909_begin_0, end = var_7909_end_0, end_mask = var_7909_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7909_cast")];
            tensor<int32, [4]> var_7913_begin_0 = const()[name = tensor<string, []>("op_7913_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_7913_end_0 = const()[name = tensor<string, []>("op_7913_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_7913_end_mask_0 = const()[name = tensor<string, []>("op_7913_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7913_cast = slice_by_index(begin = var_7913_begin_0, end = var_7913_end_0, end_mask = var_7913_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7913_cast")];
            tensor<int32, [4]> var_7917_begin_0 = const()[name = tensor<string, []>("op_7917_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_7917_end_0 = const()[name = tensor<string, []>("op_7917_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_7917_end_mask_0 = const()[name = tensor<string, []>("op_7917_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7917_cast = slice_by_index(begin = var_7917_begin_0, end = var_7917_end_0, end_mask = var_7917_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7917_cast")];
            tensor<int32, [4]> var_7921_begin_0 = const()[name = tensor<string, []>("op_7921_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_7921_end_0 = const()[name = tensor<string, []>("op_7921_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_7921_end_mask_0 = const()[name = tensor<string, []>("op_7921_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7921_cast = slice_by_index(begin = var_7921_begin_0, end = var_7921_end_0, end_mask = var_7921_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7921_cast")];
            tensor<int32, [4]> var_7925_begin_0 = const()[name = tensor<string, []>("op_7925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_7925_end_0 = const()[name = tensor<string, []>("op_7925_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_7925_end_mask_0 = const()[name = tensor<string, []>("op_7925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7925_cast = slice_by_index(begin = var_7925_begin_0, end = var_7925_end_0, end_mask = var_7925_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7925_cast")];
            tensor<int32, [4]> var_7929_begin_0 = const()[name = tensor<string, []>("op_7929_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_7929_end_0 = const()[name = tensor<string, []>("op_7929_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_7929_end_mask_0 = const()[name = tensor<string, []>("op_7929_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7929_cast = slice_by_index(begin = var_7929_begin_0, end = var_7929_end_0, end_mask = var_7929_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7929_cast")];
            tensor<int32, [4]> var_7933_begin_0 = const()[name = tensor<string, []>("op_7933_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_7933_end_0 = const()[name = tensor<string, []>("op_7933_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_7933_end_mask_0 = const()[name = tensor<string, []>("op_7933_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7933_cast = slice_by_index(begin = var_7933_begin_0, end = var_7933_end_0, end_mask = var_7933_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7933_cast")];
            tensor<int32, [4]> var_7937_begin_0 = const()[name = tensor<string, []>("op_7937_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_7937_end_0 = const()[name = tensor<string, []>("op_7937_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_7937_end_mask_0 = const()[name = tensor<string, []>("op_7937_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7937_cast = slice_by_index(begin = var_7937_begin_0, end = var_7937_end_0, end_mask = var_7937_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7937_cast")];
            tensor<int32, [4]> var_7941_begin_0 = const()[name = tensor<string, []>("op_7941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_7941_end_0 = const()[name = tensor<string, []>("op_7941_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_7941_end_mask_0 = const()[name = tensor<string, []>("op_7941_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7941_cast = slice_by_index(begin = var_7941_begin_0, end = var_7941_end_0, end_mask = var_7941_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7941_cast")];
            tensor<int32, [4]> var_7945_begin_0 = const()[name = tensor<string, []>("op_7945_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_7945_end_0 = const()[name = tensor<string, []>("op_7945_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_7945_end_mask_0 = const()[name = tensor<string, []>("op_7945_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7945_cast = slice_by_index(begin = var_7945_begin_0, end = var_7945_end_0, end_mask = var_7945_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7945_cast")];
            tensor<int32, [4]> var_7949_begin_0 = const()[name = tensor<string, []>("op_7949_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_7949_end_0 = const()[name = tensor<string, []>("op_7949_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_7949_end_mask_0 = const()[name = tensor<string, []>("op_7949_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7949_cast = slice_by_index(begin = var_7949_begin_0, end = var_7949_end_0, end_mask = var_7949_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7949_cast")];
            tensor<int32, [4]> var_7953_begin_0 = const()[name = tensor<string, []>("op_7953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_7953_end_0 = const()[name = tensor<string, []>("op_7953_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_7953_end_mask_0 = const()[name = tensor<string, []>("op_7953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_7953_cast = slice_by_index(begin = var_7953_begin_0, end = var_7953_end_0, end_mask = var_7953_end_mask_0, x = transpose_16)[name = tensor<string, []>("op_7953_cast")];
            tensor<int32, [4]> var_7955_begin_0 = const()[name = tensor<string, []>("op_7955_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_7955_end_0 = const()[name = tensor<string, []>("op_7955_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_7955_end_mask_0 = const()[name = tensor<string, []>("op_7955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7955_cast = slice_by_index(begin = var_7955_begin_0, end = var_7955_end_0, end_mask = var_7955_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7955_cast")];
            tensor<int32, [4]> var_7959_begin_0 = const()[name = tensor<string, []>("op_7959_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_7959_end_0 = const()[name = tensor<string, []>("op_7959_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_7959_end_mask_0 = const()[name = tensor<string, []>("op_7959_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7959_cast = slice_by_index(begin = var_7959_begin_0, end = var_7959_end_0, end_mask = var_7959_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7959_cast")];
            tensor<int32, [4]> var_7963_begin_0 = const()[name = tensor<string, []>("op_7963_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_7963_end_0 = const()[name = tensor<string, []>("op_7963_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_7963_end_mask_0 = const()[name = tensor<string, []>("op_7963_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7963_cast = slice_by_index(begin = var_7963_begin_0, end = var_7963_end_0, end_mask = var_7963_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7963_cast")];
            tensor<int32, [4]> var_7967_begin_0 = const()[name = tensor<string, []>("op_7967_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_7967_end_0 = const()[name = tensor<string, []>("op_7967_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_7967_end_mask_0 = const()[name = tensor<string, []>("op_7967_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7967_cast = slice_by_index(begin = var_7967_begin_0, end = var_7967_end_0, end_mask = var_7967_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7967_cast")];
            tensor<int32, [4]> var_7971_begin_0 = const()[name = tensor<string, []>("op_7971_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_7971_end_0 = const()[name = tensor<string, []>("op_7971_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_7971_end_mask_0 = const()[name = tensor<string, []>("op_7971_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7971_cast = slice_by_index(begin = var_7971_begin_0, end = var_7971_end_0, end_mask = var_7971_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7971_cast")];
            tensor<int32, [4]> var_7975_begin_0 = const()[name = tensor<string, []>("op_7975_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_7975_end_0 = const()[name = tensor<string, []>("op_7975_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_7975_end_mask_0 = const()[name = tensor<string, []>("op_7975_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7975_cast = slice_by_index(begin = var_7975_begin_0, end = var_7975_end_0, end_mask = var_7975_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7975_cast")];
            tensor<int32, [4]> var_7979_begin_0 = const()[name = tensor<string, []>("op_7979_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_7979_end_0 = const()[name = tensor<string, []>("op_7979_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_7979_end_mask_0 = const()[name = tensor<string, []>("op_7979_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7979_cast = slice_by_index(begin = var_7979_begin_0, end = var_7979_end_0, end_mask = var_7979_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7979_cast")];
            tensor<int32, [4]> var_7983_begin_0 = const()[name = tensor<string, []>("op_7983_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_7983_end_0 = const()[name = tensor<string, []>("op_7983_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_7983_end_mask_0 = const()[name = tensor<string, []>("op_7983_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7983_cast = slice_by_index(begin = var_7983_begin_0, end = var_7983_end_0, end_mask = var_7983_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7983_cast")];
            tensor<int32, [4]> var_7987_begin_0 = const()[name = tensor<string, []>("op_7987_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_7987_end_0 = const()[name = tensor<string, []>("op_7987_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_7987_end_mask_0 = const()[name = tensor<string, []>("op_7987_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7987_cast = slice_by_index(begin = var_7987_begin_0, end = var_7987_end_0, end_mask = var_7987_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7987_cast")];
            tensor<int32, [4]> var_7991_begin_0 = const()[name = tensor<string, []>("op_7991_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_7991_end_0 = const()[name = tensor<string, []>("op_7991_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_7991_end_mask_0 = const()[name = tensor<string, []>("op_7991_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7991_cast = slice_by_index(begin = var_7991_begin_0, end = var_7991_end_0, end_mask = var_7991_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7991_cast")];
            tensor<int32, [4]> var_7995_begin_0 = const()[name = tensor<string, []>("op_7995_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_7995_end_0 = const()[name = tensor<string, []>("op_7995_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_7995_end_mask_0 = const()[name = tensor<string, []>("op_7995_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_7995_cast = slice_by_index(begin = var_7995_begin_0, end = var_7995_end_0, end_mask = var_7995_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_7995_cast")];
            tensor<int32, [4]> var_7999_begin_0 = const()[name = tensor<string, []>("op_7999_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_7999_end_0 = const()[name = tensor<string, []>("op_7999_end_0"), val = tensor<int32, [4]>([2, 768, 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, 64, 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_31_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, 768, 0, 0])];
            tensor<int32, [4]> var_8003_end_0 = const()[name = tensor<string, []>("op_8003_end_0"), val = tensor<int32, [4]>([2, 832, 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, 64, 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_31_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, 832, 0, 0])];
            tensor<int32, [4]> var_8007_end_0 = const()[name = tensor<string, []>("op_8007_end_0"), val = tensor<int32, [4]>([2, 896, 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, 64, 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_31_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, 896, 0, 0])];
            tensor<int32, [4]> var_8011_end_0 = const()[name = tensor<string, []>("op_8011_end_0"), val = tensor<int32, [4]>([2, 960, 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, 64, 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_31_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, 960, 0, 0])];
            tensor<int32, [4]> var_8015_end_0 = const()[name = tensor<string, []>("op_8015_end_0"), val = tensor<int32, [4]>([2, 1024, 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, 64, 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_31_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, 1024, 0, 0])];
            tensor<int32, [4]> var_8019_end_0 = const()[name = tensor<string, []>("op_8019_end_0"), val = tensor<int32, [4]>([2, 1088, 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, 64, 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_31_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, 1088, 0, 0])];
            tensor<int32, [4]> var_8023_end_0 = const()[name = tensor<string, []>("op_8023_end_0"), val = tensor<int32, [4]>([2, 1152, 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, 64, 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_31_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, 1152, 0, 0])];
            tensor<int32, [4]> var_8027_end_0 = const()[name = tensor<string, []>("op_8027_end_0"), val = tensor<int32, [4]>([2, 1216, 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, 64, 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_31_cast)[name = tensor<string, []>("op_8027_cast")];
            tensor<int32, [4]> var_8031_begin_0 = const()[name = tensor<string, []>("op_8031_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8031_end_0 = const()[name = tensor<string, []>("op_8031_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_8031_end_mask_0 = const()[name = tensor<string, []>("op_8031_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8031_cast = slice_by_index(begin = var_8031_begin_0, end = var_8031_end_0, end_mask = var_8031_end_mask_0, x = v_31_cast)[name = tensor<string, []>("op_8031_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_7877_cast, var_7794_cast))[name = tensor<string, []>("op_8035_cast")];
            tensor<fp16, []> var_8036_to_fp16 = const()[name = tensor<string, []>("op_8036_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_281_cast = mul(x = var_8035_cast, y = var_8036_to_fp16)[name = tensor<string, []>("aw_281_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_7881_cast, var_7798_cast))[name = tensor<string, []>("op_8039_cast")];
            tensor<fp16, []> var_8040_to_fp16 = const()[name = tensor<string, []>("op_8040_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_283_cast = mul(x = var_8039_cast, y = var_8040_to_fp16)[name = tensor<string, []>("aw_283_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_7885_cast, var_7802_cast))[name = tensor<string, []>("op_8043_cast")];
            tensor<fp16, []> var_8044_to_fp16 = const()[name = tensor<string, []>("op_8044_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_285_cast = mul(x = var_8043_cast, y = var_8044_to_fp16)[name = tensor<string, []>("aw_285_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_7889_cast, var_7806_cast))[name = tensor<string, []>("op_8047_cast")];
            tensor<fp16, []> var_8048_to_fp16 = const()[name = tensor<string, []>("op_8048_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_287_cast = mul(x = var_8047_cast, y = var_8048_to_fp16)[name = tensor<string, []>("aw_287_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_7893_cast, var_7810_cast))[name = tensor<string, []>("op_8051_cast")];
            tensor<fp16, []> var_8052_to_fp16 = const()[name = tensor<string, []>("op_8052_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_289_cast = mul(x = var_8051_cast, y = var_8052_to_fp16)[name = tensor<string, []>("aw_289_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_7897_cast, var_7814_cast))[name = tensor<string, []>("op_8055_cast")];
            tensor<fp16, []> var_8056_to_fp16 = const()[name = tensor<string, []>("op_8056_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_291_cast = mul(x = var_8055_cast, y = var_8056_to_fp16)[name = tensor<string, []>("aw_291_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_7901_cast, var_7818_cast))[name = tensor<string, []>("op_8059_cast")];
            tensor<fp16, []> var_8060_to_fp16 = const()[name = tensor<string, []>("op_8060_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_293_cast = mul(x = var_8059_cast, y = var_8060_to_fp16)[name = tensor<string, []>("aw_293_cast")];
            tensor<string, []> var_8063_equation_0 = const()[name = tensor<string, []>("op_8063_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8063_cast = einsum(equation = var_8063_equation_0, values = (var_7905_cast, var_7822_cast))[name = tensor<string, []>("op_8063_cast")];
            tensor<fp16, []> var_8064_to_fp16 = const()[name = tensor<string, []>("op_8064_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_295_cast = mul(x = var_8063_cast, y = var_8064_to_fp16)[name = tensor<string, []>("aw_295_cast")];
            tensor<string, []> var_8067_equation_0 = const()[name = tensor<string, []>("op_8067_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8067_cast = einsum(equation = var_8067_equation_0, values = (var_7909_cast, var_7826_cast))[name = tensor<string, []>("op_8067_cast")];
            tensor<fp16, []> var_8068_to_fp16 = const()[name = tensor<string, []>("op_8068_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_297_cast = mul(x = var_8067_cast, y = var_8068_to_fp16)[name = tensor<string, []>("aw_297_cast")];
            tensor<string, []> var_8071_equation_0 = const()[name = tensor<string, []>("op_8071_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8071_cast = einsum(equation = var_8071_equation_0, values = (var_7913_cast, var_7830_cast))[name = tensor<string, []>("op_8071_cast")];
            tensor<fp16, []> var_8072_to_fp16 = const()[name = tensor<string, []>("op_8072_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_299_cast = mul(x = var_8071_cast, y = var_8072_to_fp16)[name = tensor<string, []>("aw_299_cast")];
            tensor<string, []> var_8075_equation_0 = const()[name = tensor<string, []>("op_8075_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8075_cast = einsum(equation = var_8075_equation_0, values = (var_7917_cast, var_7834_cast))[name = tensor<string, []>("op_8075_cast")];
            tensor<fp16, []> var_8076_to_fp16 = const()[name = tensor<string, []>("op_8076_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_301_cast = mul(x = var_8075_cast, y = var_8076_to_fp16)[name = tensor<string, []>("aw_301_cast")];
            tensor<string, []> var_8079_equation_0 = const()[name = tensor<string, []>("op_8079_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8079_cast = einsum(equation = var_8079_equation_0, values = (var_7921_cast, var_7838_cast))[name = tensor<string, []>("op_8079_cast")];
            tensor<fp16, []> var_8080_to_fp16 = const()[name = tensor<string, []>("op_8080_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_303_cast = mul(x = var_8079_cast, y = var_8080_to_fp16)[name = tensor<string, []>("aw_303_cast")];
            tensor<string, []> var_8083_equation_0 = const()[name = tensor<string, []>("op_8083_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8083_cast = einsum(equation = var_8083_equation_0, values = (var_7925_cast, var_7842_cast))[name = tensor<string, []>("op_8083_cast")];
            tensor<fp16, []> var_8084_to_fp16 = const()[name = tensor<string, []>("op_8084_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_305_cast = mul(x = var_8083_cast, y = var_8084_to_fp16)[name = tensor<string, []>("aw_305_cast")];
            tensor<string, []> var_8087_equation_0 = const()[name = tensor<string, []>("op_8087_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8087_cast = einsum(equation = var_8087_equation_0, values = (var_7929_cast, var_7846_cast))[name = tensor<string, []>("op_8087_cast")];
            tensor<fp16, []> var_8088_to_fp16 = const()[name = tensor<string, []>("op_8088_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_307_cast = mul(x = var_8087_cast, y = var_8088_to_fp16)[name = tensor<string, []>("aw_307_cast")];
            tensor<string, []> var_8091_equation_0 = const()[name = tensor<string, []>("op_8091_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8091_cast = einsum(equation = var_8091_equation_0, values = (var_7933_cast, var_7850_cast))[name = tensor<string, []>("op_8091_cast")];
            tensor<fp16, []> var_8092_to_fp16 = const()[name = tensor<string, []>("op_8092_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_309_cast = mul(x = var_8091_cast, y = var_8092_to_fp16)[name = tensor<string, []>("aw_309_cast")];
            tensor<string, []> var_8095_equation_0 = const()[name = tensor<string, []>("op_8095_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8095_cast = einsum(equation = var_8095_equation_0, values = (var_7937_cast, var_7854_cast))[name = tensor<string, []>("op_8095_cast")];
            tensor<fp16, []> var_8096_to_fp16 = const()[name = tensor<string, []>("op_8096_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_311_cast = mul(x = var_8095_cast, y = var_8096_to_fp16)[name = tensor<string, []>("aw_311_cast")];
            tensor<string, []> var_8099_equation_0 = const()[name = tensor<string, []>("op_8099_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8099_cast = einsum(equation = var_8099_equation_0, values = (var_7941_cast, var_7858_cast))[name = tensor<string, []>("op_8099_cast")];
            tensor<fp16, []> var_8100_to_fp16 = const()[name = tensor<string, []>("op_8100_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_313_cast = mul(x = var_8099_cast, y = var_8100_to_fp16)[name = tensor<string, []>("aw_313_cast")];
            tensor<string, []> var_8103_equation_0 = const()[name = tensor<string, []>("op_8103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8103_cast = einsum(equation = var_8103_equation_0, values = (var_7945_cast, var_7862_cast))[name = tensor<string, []>("op_8103_cast")];
            tensor<fp16, []> var_8104_to_fp16 = const()[name = tensor<string, []>("op_8104_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_315_cast = mul(x = var_8103_cast, y = var_8104_to_fp16)[name = tensor<string, []>("aw_315_cast")];
            tensor<string, []> var_8107_equation_0 = const()[name = tensor<string, []>("op_8107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8107_cast = einsum(equation = var_8107_equation_0, values = (var_7949_cast, var_7866_cast))[name = tensor<string, []>("op_8107_cast")];
            tensor<fp16, []> var_8108_to_fp16 = const()[name = tensor<string, []>("op_8108_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_317_cast = mul(x = var_8107_cast, y = var_8108_to_fp16)[name = tensor<string, []>("aw_317_cast")];
            tensor<string, []> var_8111_equation_0 = const()[name = tensor<string, []>("op_8111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_8111_cast = einsum(equation = var_8111_equation_0, values = (var_7953_cast, var_7870_cast))[name = tensor<string, []>("op_8111_cast")];
            tensor<fp16, []> var_8112_to_fp16 = const()[name = tensor<string, []>("op_8112_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_319_cast = mul(x = var_8111_cast, y = var_8112_to_fp16)[name = tensor<string, []>("aw_319_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8114_cast = softmax(axis = var_7228, x = aw_281_cast)[name = tensor<string, []>("op_8114_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8115_cast = softmax(axis = var_7228, x = aw_283_cast)[name = tensor<string, []>("op_8115_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8116_cast = softmax(axis = var_7228, x = aw_285_cast)[name = tensor<string, []>("op_8116_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8117_cast = softmax(axis = var_7228, x = aw_287_cast)[name = tensor<string, []>("op_8117_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8118_cast = softmax(axis = var_7228, x = aw_289_cast)[name = tensor<string, []>("op_8118_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8119_cast = softmax(axis = var_7228, x = aw_291_cast)[name = tensor<string, []>("op_8119_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8120_cast = softmax(axis = var_7228, x = aw_293_cast)[name = tensor<string, []>("op_8120_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8121_cast = softmax(axis = var_7228, x = aw_295_cast)[name = tensor<string, []>("op_8121_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8122_cast = softmax(axis = var_7228, x = aw_297_cast)[name = tensor<string, []>("op_8122_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8123_cast = softmax(axis = var_7228, x = aw_299_cast)[name = tensor<string, []>("op_8123_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8124_cast = softmax(axis = var_7228, x = aw_301_cast)[name = tensor<string, []>("op_8124_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8125_cast = softmax(axis = var_7228, x = aw_303_cast)[name = tensor<string, []>("op_8125_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8126_cast = softmax(axis = var_7228, x = aw_305_cast)[name = tensor<string, []>("op_8126_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8127_cast = softmax(axis = var_7228, x = aw_307_cast)[name = tensor<string, []>("op_8127_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8128_cast = softmax(axis = var_7228, x = aw_309_cast)[name = tensor<string, []>("op_8128_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8129_cast = softmax(axis = var_7228, x = aw_311_cast)[name = tensor<string, []>("op_8129_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8130_cast = softmax(axis = var_7228, x = aw_313_cast)[name = tensor<string, []>("op_8130_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8131_cast = softmax(axis = var_7228, x = aw_315_cast)[name = tensor<string, []>("op_8131_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8132_cast = softmax(axis = var_7228, x = aw_317_cast)[name = tensor<string, []>("op_8132_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_8133_cast = softmax(axis = var_7228, x = aw_319_cast)[name = tensor<string, []>("op_8133_cast")];
            tensor<string, []> var_8135_equation_0 = const()[name = tensor<string, []>("op_8135_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8135_cast = einsum(equation = var_8135_equation_0, values = (var_7955_cast, var_8114_cast))[name = tensor<string, []>("op_8135_cast")];
            tensor<string, []> var_8137_equation_0 = const()[name = tensor<string, []>("op_8137_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8137_cast = einsum(equation = var_8137_equation_0, values = (var_7959_cast, var_8115_cast))[name = tensor<string, []>("op_8137_cast")];
            tensor<string, []> var_8139_equation_0 = const()[name = tensor<string, []>("op_8139_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8139_cast = einsum(equation = var_8139_equation_0, values = (var_7963_cast, var_8116_cast))[name = tensor<string, []>("op_8139_cast")];
            tensor<string, []> var_8141_equation_0 = const()[name = tensor<string, []>("op_8141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8141_cast = einsum(equation = var_8141_equation_0, values = (var_7967_cast, var_8117_cast))[name = tensor<string, []>("op_8141_cast")];
            tensor<string, []> var_8143_equation_0 = const()[name = tensor<string, []>("op_8143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8143_cast = einsum(equation = var_8143_equation_0, values = (var_7971_cast, var_8118_cast))[name = tensor<string, []>("op_8143_cast")];
            tensor<string, []> var_8145_equation_0 = const()[name = tensor<string, []>("op_8145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8145_cast = einsum(equation = var_8145_equation_0, values = (var_7975_cast, var_8119_cast))[name = tensor<string, []>("op_8145_cast")];
            tensor<string, []> var_8147_equation_0 = const()[name = tensor<string, []>("op_8147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8147_cast = einsum(equation = var_8147_equation_0, values = (var_7979_cast, var_8120_cast))[name = tensor<string, []>("op_8147_cast")];
            tensor<string, []> var_8149_equation_0 = const()[name = tensor<string, []>("op_8149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8149_cast = einsum(equation = var_8149_equation_0, values = (var_7983_cast, var_8121_cast))[name = tensor<string, []>("op_8149_cast")];
            tensor<string, []> var_8151_equation_0 = const()[name = tensor<string, []>("op_8151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8151_cast = einsum(equation = var_8151_equation_0, values = (var_7987_cast, var_8122_cast))[name = tensor<string, []>("op_8151_cast")];
            tensor<string, []> var_8153_equation_0 = const()[name = tensor<string, []>("op_8153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8153_cast = einsum(equation = var_8153_equation_0, values = (var_7991_cast, var_8123_cast))[name = tensor<string, []>("op_8153_cast")];
            tensor<string, []> var_8155_equation_0 = const()[name = tensor<string, []>("op_8155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8155_cast = einsum(equation = var_8155_equation_0, values = (var_7995_cast, var_8124_cast))[name = tensor<string, []>("op_8155_cast")];
            tensor<string, []> var_8157_equation_0 = const()[name = tensor<string, []>("op_8157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8157_cast = einsum(equation = var_8157_equation_0, values = (var_7999_cast, var_8125_cast))[name = tensor<string, []>("op_8157_cast")];
            tensor<string, []> var_8159_equation_0 = const()[name = tensor<string, []>("op_8159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8159_cast = einsum(equation = var_8159_equation_0, values = (var_8003_cast, var_8126_cast))[name = tensor<string, []>("op_8159_cast")];
            tensor<string, []> var_8161_equation_0 = const()[name = tensor<string, []>("op_8161_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8161_cast = einsum(equation = var_8161_equation_0, values = (var_8007_cast, var_8127_cast))[name = tensor<string, []>("op_8161_cast")];
            tensor<string, []> var_8163_equation_0 = const()[name = tensor<string, []>("op_8163_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8163_cast = einsum(equation = var_8163_equation_0, values = (var_8011_cast, var_8128_cast))[name = tensor<string, []>("op_8163_cast")];
            tensor<string, []> var_8165_equation_0 = const()[name = tensor<string, []>("op_8165_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8165_cast = einsum(equation = var_8165_equation_0, values = (var_8015_cast, var_8129_cast))[name = tensor<string, []>("op_8165_cast")];
            tensor<string, []> var_8167_equation_0 = const()[name = tensor<string, []>("op_8167_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8167_cast = einsum(equation = var_8167_equation_0, values = (var_8019_cast, var_8130_cast))[name = tensor<string, []>("op_8167_cast")];
            tensor<string, []> var_8169_equation_0 = const()[name = tensor<string, []>("op_8169_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8169_cast = einsum(equation = var_8169_equation_0, values = (var_8023_cast, var_8131_cast))[name = tensor<string, []>("op_8169_cast")];
            tensor<string, []> var_8171_equation_0 = const()[name = tensor<string, []>("op_8171_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8171_cast = einsum(equation = var_8171_equation_0, values = (var_8027_cast, var_8132_cast))[name = tensor<string, []>("op_8171_cast")];
            tensor<string, []> var_8173_equation_0 = const()[name = tensor<string, []>("op_8173_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8173_cast = einsum(equation = var_8173_equation_0, values = (var_8031_cast, var_8133_cast))[name = tensor<string, []>("op_8173_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_7228, interleave = input_303_interleave_0, values = (var_8135_cast, var_8137_cast, var_8139_cast, var_8141_cast, var_8143_cast, var_8145_cast, var_8147_cast, var_8149_cast, var_8151_cast, var_8153_cast, var_8155_cast, var_8157_cast, var_8159_cast, var_8161_cast, var_8163_cast, var_8165_cast, var_8167_cast, var_8169_cast, var_8171_cast, var_8173_cast))[name = tensor<string, []>("input_303_cast")];
            tensor<int32, [2]> var_8179 = const()[name = tensor<string, []>("op_8179"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8181 = const()[name = tensor<string, []>("op_8181"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8183_pad_type_0 = const()[name = tensor<string, []>("op_8183_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8183_pad_0 = const()[name = tensor<string, []>("op_8183_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144248576))), lut = tensor<fp16, [4]>([-0x1.3d8p-5, -0x1.77p-7, 0x1.7b8p-7, 0x1.3e4p-5]), 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, []>(144658240)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8183_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_8181, groups = var_7228, pad = var_8183_pad_0, pad_type = var_8183_pad_type_0, strides = var_8179, 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_8183_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_47_cast = add(x = var_8183_cast, y = inputs_45_cast)[name = tensor<string, []>("inputs_47_cast")];
            tensor<int32, [1]> var_8187 = const()[name = tensor<string, []>("op_8187"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_47_cast = reduce_mean(axes = var_8187, keep_dims = var_7223, 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_8191 = const()[name = tensor<string, []>("op_8191"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_8192_cast = reduce_mean(axes = var_8191, keep_dims = var_7223, x = zero_mean_sq_47_cast)[name = tensor<string, []>("op_8192_cast")];
            tensor<fp16, []> var_8193_to_fp16 = const()[name = tensor<string, []>("op_8193_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_8194_cast = add(x = var_8192_cast, y = var_8193_to_fp16)[name = tensor<string, []>("op_8194_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_8194_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_8198_to_fp16 = const()[name = tensor<string, []>("op_8198_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144660864)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8199_cast = add(x = out_47_cast, y = var_8198_to_fp16)[name = tensor<string, []>("op_8199_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8201_to_fp16 = const()[name = tensor<string, []>("op_8201_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144663488)))];
            tensor<fp16, [2, 1280, 1, 256]> input_305_cast = mul(x = var_8199_cast, y = var_8201_to_fp16)[name = tensor<string, []>("input_305_cast")];
            tensor<int32, [2]> var_8209 = const()[name = tensor<string, []>("op_8209"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8211 = const()[name = tensor<string, []>("op_8211"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8213_pad_type_0 = const()[name = tensor<string, []>("op_8213_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8213_pad_0 = const()[name = tensor<string, []>("op_8213_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, [3276800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144666112))), lut = tensor<fp16, [4]>([-0x1.58p-4, -0x1.974p-6, 0x1.95cp-6, 0x1.578p-4]), 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, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147942976))), lut = tensor<fp16, [4]>([-0x1.584p-3, -0x1.064p-4, -0x1.264p-8, 0x1.f9p-5]), 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_8213_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_8211, groups = var_7228, pad = var_8213_pad_0, pad_type = var_8213_pad_type_0, strides = var_8209, 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_8213_cast")];
            tensor<int32, [2]> var_8214_split_sizes_0 = const()[name = tensor<string, []>("op_8214_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_8214_axis_0 = const()[name = tensor<string, []>("op_8214_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_8214_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_8214_cast_1 = split(axis = var_8214_axis_0, split_sizes = var_8214_split_sizes_0, x = var_8213_cast)[name = tensor<string, []>("op_8214_cast")];
            tensor<string, []> var_8216_mode_0 = const()[name = tensor<string, []>("op_8216_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_8216_cast = gelu(mode = var_8216_mode_0, x = var_8214_cast_1)[name = tensor<string, []>("op_8216_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_307_cast = mul(x = var_8214_cast_0, y = var_8216_cast)[name = tensor<string, []>("input_307_cast")];
            tensor<int32, [2]> var_8220 = const()[name = tensor<string, []>("op_8220"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8222 = const()[name = tensor<string, []>("op_8222"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8224_pad_type_0 = const()[name = tensor<string, []>("op_8224_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8224_pad_0 = const()[name = tensor<string, []>("op_8224_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, [1638400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147945600))), lut = tensor<fp16, [4]>([-0x1.56cp-4, -0x1.99cp-6, 0x1.994p-6, 0x1.568p-4]), 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, []>(149584064)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8224_cast = conv(bias = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_8222, groups = var_7228, pad = var_8224_pad_0, pad_type = var_8224_pad_type_0, strides = var_8220, weight = up_blocks_1_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_307_cast)[name = tensor<string, []>("op_8224_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_173_cast = add(x = var_8224_cast, y = inputs_47_cast)[name = tensor<string, []>("hidden_states_173_cast")];
            tensor<int32, [4]> var_8226 = const()[name = tensor<string, []>("op_8226"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_309_cast = reshape(shape = var_8226, x = hidden_states_173_cast)[name = tensor<string, []>("input_309_cast")];
            tensor<int32, [2]> var_8230 = const()[name = tensor<string, []>("op_8230"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8232 = const()[name = tensor<string, []>("op_8232"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149586688))), lut = tensor<fp16, [4]>([-0x1.56cp-4, -0x1.9acp-6, 0x1.9bcp-6, 0x1.578p-4]), 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, []>(149996352)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_175_cast = conv(bias = up_blocks_1_attentions_0_proj_out_bias_to_fp16, dilations = var_8232, groups = var_7228, pad = hidden_states_175_pad_0, pad_type = hidden_states_175_pad_type_0, strides = var_8230, 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_7228, 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149998976))), lut = tensor<fp16, [4]>([0x1.a54p-3, 0x1.0acp-1, 0x1.6d4p-2, 0x1.798p-1]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149999680))), lut = tensor<fp16, [4]>([-0x1.d98p-6, -0x1.b38p-2, -0x1.8ecp-3, -0x1.6fcp-1]), 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_8250 = const()[name = tensor<string, []>("op_8250"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8252 = const()[name = tensor<string, []>("op_8252"), 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, [7372800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150000384))), lut = tensor<fp16, [4]>([-0x1.ffp-5, -0x1.218p-6, 0x1.29p-6, 0x1.04p-4]), 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, []>(157373248)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_179_cast = conv(bias = up_blocks_1_resnets_1_conv1_bias_to_fp16, dilations = var_8252, groups = var_7228, pad = hidden_states_179_pad_0, pad_type = hidden_states_179_pad_type_0, strides = var_8250, 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_8258 = const()[name = tensor<string, []>("op_8258"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8260 = const()[name = tensor<string, []>("op_8260"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157375872))), lut = tensor<fp16, [4]>([-0x1.b84p-5, -0x1.a98p-9, 0x1.a5p-9, 0x1.79p-5]), 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, []>(157785536)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_29_cast = conv(bias = up_blocks_1_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_8260, groups = var_7228, pad = temb_29_pad_0, pad_type = temb_29_pad_type_0, strides = var_8258, 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, []>(157788160)))];
            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, []>(157790784)))];
            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_8270 = const()[name = tensor<string, []>("op_8270"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8272 = const()[name = tensor<string, []>("op_8272"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157793408))), lut = tensor<fp16, [4]>([-0x1.3ap-4, -0x1.654p-6, 0x1.69cp-6, 0x1.3c4p-4]), 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, []>(161479872)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_181_cast = conv(bias = up_blocks_1_resnets_1_conv2_bias_to_fp16, dilations = var_8272, groups = var_7228, pad = hidden_states_181_pad_0, pad_type = hidden_states_181_pad_type_0, strides = var_8270, 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_8277 = const()[name = tensor<string, []>("op_8277"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8279 = const()[name = tensor<string, []>("op_8279"), 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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161482496))), lut = tensor<fp16, [4]>([-0x1.e7cp-6, -0x1.1bcp-7, 0x1.29cp-7, 0x1.f08p-6]), 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, []>(162301760)))];
            tensor<fp16, [2, 1280, 16, 16]> x_13_cast = conv(bias = up_blocks_1_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_8279, groups = var_7228, pad = x_13_pad_0, pad_type = x_13_pad_type_0, strides = var_8277, 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, []>(162304384)))];
            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, []>(162307008)))];
            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_8299 = const()[name = tensor<string, []>("op_8299"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8301 = const()[name = tensor<string, []>("op_8301"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162309632))), lut = tensor<fp16, [4]>([-0x1.9bp-4, -0x1.eccp-6, 0x1.eb4p-6, 0x1.9a8p-4]), 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, []>(162719296)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_185_cast = conv(bias = up_blocks_1_attentions_1_proj_in_bias_to_fp16, dilations = var_8301, groups = var_7228, pad = hidden_states_185_pad_0, pad_type = hidden_states_185_pad_type_0, strides = var_8299, 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_8306 = const()[name = tensor<string, []>("op_8306"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_49_cast = reshape(shape = var_8306, x = hidden_states_185_cast)[name = tensor<string, []>("inputs_49_cast")];
            tensor<int32, [1]> var_8316 = const()[name = tensor<string, []>("op_8316"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_49_cast = reduce_mean(axes = var_8316, keep_dims = var_7223, 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_8320 = const()[name = tensor<string, []>("op_8320"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_8321_cast = reduce_mean(axes = var_8320, keep_dims = var_7223, x = zero_mean_sq_49_cast)[name = tensor<string, []>("op_8321_cast")];
            tensor<fp16, []> var_8322_to_fp16 = const()[name = tensor<string, []>("op_8322_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_8323_cast = add(x = var_8321_cast, y = var_8322_to_fp16)[name = tensor<string, []>("op_8323_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_8323_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_8327_to_fp16 = const()[name = tensor<string, []>("op_8327_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162721920)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8328_cast = add(x = out_49_cast, y = var_8327_to_fp16)[name = tensor<string, []>("op_8328_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8330_to_fp16 = const()[name = tensor<string, []>("op_8330_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162724544)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_187_cast = mul(x = var_8328_cast, y = var_8330_to_fp16)[name = tensor<string, []>("hidden_states_187_cast")];
            tensor<int32, [2]> var_8337 = const()[name = tensor<string, []>("op_8337"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8339 = const()[name = tensor<string, []>("op_8339"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162727168))), lut = tensor<fp16, [4]>([-0x1.7ep-4, -0x1.ca4p-6, 0x1.c5cp-6, 0x1.7d4p-4]), 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_8339, groups = var_7228, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = var_8337, 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_8343 = const()[name = tensor<string, []>("op_8343"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8345 = const()[name = tensor<string, []>("op_8345"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163136832))), lut = tensor<fp16, [4]>([-0x1.7f8p-4, -0x1.c7p-6, 0x1.cb4p-6, 0x1.808p-4]), 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_8345, groups = var_7228, pad = k_65_pad_0, pad_type = k_65_pad_type_0, strides = var_8343, 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_8349 = const()[name = tensor<string, []>("op_8349"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8351 = const()[name = tensor<string, []>("op_8351"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163546496))), lut = tensor<fp16, [4]>([-0x1.52p-4, -0x1.93p-6, 0x1.944p-6, 0x1.52p-4]), 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_8351, groups = var_7228, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = var_8349, 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_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, 64, 1, 256])];
            tensor<bool, [4]> var_8355_end_mask_0 = const()[name = tensor<string, []>("op_8355_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8355_cast = slice_by_index(begin = var_8355_begin_0, end = var_8355_end_0, end_mask = var_8355_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8355_cast")];
            tensor<int32, [4]> var_8359_begin_0 = const()[name = tensor<string, []>("op_8359_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8359_end_0 = const()[name = tensor<string, []>("op_8359_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_8359_end_mask_0 = const()[name = tensor<string, []>("op_8359_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8359_cast = slice_by_index(begin = var_8359_begin_0, end = var_8359_end_0, end_mask = var_8359_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8359_cast")];
            tensor<int32, [4]> var_8363_begin_0 = const()[name = tensor<string, []>("op_8363_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8363_end_0 = const()[name = tensor<string, []>("op_8363_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_8363_end_mask_0 = const()[name = tensor<string, []>("op_8363_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8363_cast = slice_by_index(begin = var_8363_begin_0, end = var_8363_end_0, end_mask = var_8363_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8363_cast")];
            tensor<int32, [4]> var_8367_begin_0 = const()[name = tensor<string, []>("op_8367_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8367_end_0 = const()[name = tensor<string, []>("op_8367_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_8367_end_mask_0 = const()[name = tensor<string, []>("op_8367_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8367_cast = slice_by_index(begin = var_8367_begin_0, end = var_8367_end_0, end_mask = var_8367_end_mask_0, x = q_33_cast)[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, 256, 0, 0])];
            tensor<int32, [4]> var_8371_end_0 = const()[name = tensor<string, []>("op_8371_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_8371_end_mask_0 = const()[name = tensor<string, []>("op_8371_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8371_cast = slice_by_index(begin = var_8371_begin_0, end = var_8371_end_0, end_mask = var_8371_end_mask_0, x = q_33_cast)[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, 320, 0, 0])];
            tensor<int32, [4]> var_8375_end_0 = const()[name = tensor<string, []>("op_8375_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_8375_end_mask_0 = const()[name = tensor<string, []>("op_8375_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8375_cast = slice_by_index(begin = var_8375_begin_0, end = var_8375_end_0, end_mask = var_8375_end_mask_0, x = q_33_cast)[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, 384, 0, 0])];
            tensor<int32, [4]> var_8379_end_0 = const()[name = tensor<string, []>("op_8379_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_8379_end_mask_0 = const()[name = tensor<string, []>("op_8379_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8379_cast = slice_by_index(begin = var_8379_begin_0, end = var_8379_end_0, end_mask = var_8379_end_mask_0, x = q_33_cast)[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, 448, 0, 0])];
            tensor<int32, [4]> var_8383_end_0 = const()[name = tensor<string, []>("op_8383_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_8383_end_mask_0 = const()[name = tensor<string, []>("op_8383_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8383_cast = slice_by_index(begin = var_8383_begin_0, end = var_8383_end_0, end_mask = var_8383_end_mask_0, x = q_33_cast)[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, 512, 0, 0])];
            tensor<int32, [4]> var_8387_end_0 = const()[name = tensor<string, []>("op_8387_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_8387_end_mask_0 = const()[name = tensor<string, []>("op_8387_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8387_cast = slice_by_index(begin = var_8387_begin_0, end = var_8387_end_0, end_mask = var_8387_end_mask_0, x = q_33_cast)[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, 576, 0, 0])];
            tensor<int32, [4]> var_8391_end_0 = const()[name = tensor<string, []>("op_8391_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_8391_end_mask_0 = const()[name = tensor<string, []>("op_8391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8391_cast = slice_by_index(begin = var_8391_begin_0, end = var_8391_end_0, end_mask = var_8391_end_mask_0, x = q_33_cast)[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, 640, 0, 0])];
            tensor<int32, [4]> var_8395_end_0 = const()[name = tensor<string, []>("op_8395_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_8395_end_mask_0 = const()[name = tensor<string, []>("op_8395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8395_cast = slice_by_index(begin = var_8395_begin_0, end = var_8395_end_0, end_mask = var_8395_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8395_cast")];
            tensor<int32, [4]> var_8399_begin_0 = const()[name = tensor<string, []>("op_8399_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8399_end_0 = const()[name = tensor<string, []>("op_8399_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_8399_end_mask_0 = const()[name = tensor<string, []>("op_8399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8399_cast = slice_by_index(begin = var_8399_begin_0, end = var_8399_end_0, end_mask = var_8399_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8399_cast")];
            tensor<int32, [4]> var_8403_begin_0 = const()[name = tensor<string, []>("op_8403_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8403_end_0 = const()[name = tensor<string, []>("op_8403_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_8403_end_mask_0 = const()[name = tensor<string, []>("op_8403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8403_cast = slice_by_index(begin = var_8403_begin_0, end = var_8403_end_0, end_mask = var_8403_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8403_cast")];
            tensor<int32, [4]> var_8407_begin_0 = const()[name = tensor<string, []>("op_8407_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8407_end_0 = const()[name = tensor<string, []>("op_8407_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_8407_end_mask_0 = const()[name = tensor<string, []>("op_8407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8407_cast = slice_by_index(begin = var_8407_begin_0, end = var_8407_end_0, end_mask = var_8407_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8407_cast")];
            tensor<int32, [4]> var_8411_begin_0 = const()[name = tensor<string, []>("op_8411_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8411_end_0 = const()[name = tensor<string, []>("op_8411_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_8411_end_mask_0 = const()[name = tensor<string, []>("op_8411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8411_cast = slice_by_index(begin = var_8411_begin_0, end = var_8411_end_0, end_mask = var_8411_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8411_cast")];
            tensor<int32, [4]> var_8415_begin_0 = const()[name = tensor<string, []>("op_8415_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8415_end_0 = const()[name = tensor<string, []>("op_8415_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_8415_end_mask_0 = const()[name = tensor<string, []>("op_8415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8415_cast = slice_by_index(begin = var_8415_begin_0, end = var_8415_end_0, end_mask = var_8415_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8415_cast")];
            tensor<int32, [4]> var_8419_begin_0 = const()[name = tensor<string, []>("op_8419_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8419_end_0 = const()[name = tensor<string, []>("op_8419_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_8419_end_mask_0 = const()[name = tensor<string, []>("op_8419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8419_cast = slice_by_index(begin = var_8419_begin_0, end = var_8419_end_0, end_mask = var_8419_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8419_cast")];
            tensor<int32, [4]> var_8423_begin_0 = const()[name = tensor<string, []>("op_8423_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8423_end_0 = const()[name = tensor<string, []>("op_8423_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_8423_end_mask_0 = const()[name = tensor<string, []>("op_8423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8423_cast = slice_by_index(begin = var_8423_begin_0, end = var_8423_end_0, end_mask = var_8423_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8423_cast")];
            tensor<int32, [4]> var_8427_begin_0 = const()[name = tensor<string, []>("op_8427_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8427_end_0 = const()[name = tensor<string, []>("op_8427_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_8427_end_mask_0 = const()[name = tensor<string, []>("op_8427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8427_cast = slice_by_index(begin = var_8427_begin_0, end = var_8427_end_0, end_mask = var_8427_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8427_cast")];
            tensor<int32, [4]> var_8431_begin_0 = const()[name = tensor<string, []>("op_8431_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8431_end_0 = const()[name = tensor<string, []>("op_8431_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_8431_end_mask_0 = const()[name = tensor<string, []>("op_8431_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8431_cast = slice_by_index(begin = var_8431_begin_0, end = var_8431_end_0, end_mask = var_8431_end_mask_0, x = q_33_cast)[name = tensor<string, []>("op_8431_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_8438_begin_0 = const()[name = tensor<string, []>("op_8438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8438_end_0 = const()[name = tensor<string, []>("op_8438_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_8438_end_mask_0 = const()[name = tensor<string, []>("op_8438_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, 64]> var_8438_cast = slice_by_index(begin = var_8438_begin_0, end = var_8438_end_0, end_mask = var_8438_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8438_cast")];
            tensor<int32, [4]> var_8442_begin_0 = const()[name = tensor<string, []>("op_8442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_8442_end_0 = const()[name = tensor<string, []>("op_8442_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_8442_end_mask_0 = const()[name = tensor<string, []>("op_8442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8442_cast = slice_by_index(begin = var_8442_begin_0, end = var_8442_end_0, end_mask = var_8442_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8442_cast")];
            tensor<int32, [4]> var_8446_begin_0 = const()[name = tensor<string, []>("op_8446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_8446_end_0 = const()[name = tensor<string, []>("op_8446_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_8446_end_mask_0 = const()[name = tensor<string, []>("op_8446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8446_cast = slice_by_index(begin = var_8446_begin_0, end = var_8446_end_0, end_mask = var_8446_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8446_cast")];
            tensor<int32, [4]> var_8450_begin_0 = const()[name = tensor<string, []>("op_8450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_8450_end_0 = const()[name = tensor<string, []>("op_8450_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_8450_end_mask_0 = const()[name = tensor<string, []>("op_8450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8450_cast = slice_by_index(begin = var_8450_begin_0, end = var_8450_end_0, end_mask = var_8450_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8450_cast")];
            tensor<int32, [4]> var_8454_begin_0 = const()[name = tensor<string, []>("op_8454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_8454_end_0 = const()[name = tensor<string, []>("op_8454_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_8454_end_mask_0 = const()[name = tensor<string, []>("op_8454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8454_cast = slice_by_index(begin = var_8454_begin_0, end = var_8454_end_0, end_mask = var_8454_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8454_cast")];
            tensor<int32, [4]> var_8458_begin_0 = const()[name = tensor<string, []>("op_8458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8458_end_0 = const()[name = tensor<string, []>("op_8458_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_8458_end_mask_0 = const()[name = tensor<string, []>("op_8458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8458_cast = slice_by_index(begin = var_8458_begin_0, end = var_8458_end_0, end_mask = var_8458_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8458_cast")];
            tensor<int32, [4]> var_8462_begin_0 = const()[name = tensor<string, []>("op_8462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_8462_end_0 = const()[name = tensor<string, []>("op_8462_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_8462_end_mask_0 = const()[name = tensor<string, []>("op_8462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8462_cast = slice_by_index(begin = var_8462_begin_0, end = var_8462_end_0, end_mask = var_8462_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8462_cast")];
            tensor<int32, [4]> var_8466_begin_0 = const()[name = tensor<string, []>("op_8466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_8466_end_0 = const()[name = tensor<string, []>("op_8466_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_8466_end_mask_0 = const()[name = tensor<string, []>("op_8466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8466_cast = slice_by_index(begin = var_8466_begin_0, end = var_8466_end_0, end_mask = var_8466_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8466_cast")];
            tensor<int32, [4]> var_8470_begin_0 = const()[name = tensor<string, []>("op_8470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8470_end_0 = const()[name = tensor<string, []>("op_8470_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_8470_end_mask_0 = const()[name = tensor<string, []>("op_8470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8470_cast = slice_by_index(begin = var_8470_begin_0, end = var_8470_end_0, end_mask = var_8470_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8470_cast")];
            tensor<int32, [4]> var_8474_begin_0 = const()[name = tensor<string, []>("op_8474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_8474_end_0 = const()[name = tensor<string, []>("op_8474_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_8474_end_mask_0 = const()[name = tensor<string, []>("op_8474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8474_cast = slice_by_index(begin = var_8474_begin_0, end = var_8474_end_0, end_mask = var_8474_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8474_cast")];
            tensor<int32, [4]> var_8478_begin_0 = const()[name = tensor<string, []>("op_8478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_8478_end_0 = const()[name = tensor<string, []>("op_8478_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_8478_end_mask_0 = const()[name = tensor<string, []>("op_8478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8478_cast = slice_by_index(begin = var_8478_begin_0, end = var_8478_end_0, end_mask = var_8478_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8478_cast")];
            tensor<int32, [4]> var_8482_begin_0 = const()[name = tensor<string, []>("op_8482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_8482_end_0 = const()[name = tensor<string, []>("op_8482_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_8482_end_mask_0 = const()[name = tensor<string, []>("op_8482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8482_cast = slice_by_index(begin = var_8482_begin_0, end = var_8482_end_0, end_mask = var_8482_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8482_cast")];
            tensor<int32, [4]> var_8486_begin_0 = const()[name = tensor<string, []>("op_8486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_8486_end_0 = const()[name = tensor<string, []>("op_8486_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_8486_end_mask_0 = const()[name = tensor<string, []>("op_8486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8486_cast = slice_by_index(begin = var_8486_begin_0, end = var_8486_end_0, end_mask = var_8486_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8486_cast")];
            tensor<int32, [4]> var_8490_begin_0 = const()[name = tensor<string, []>("op_8490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_8490_end_0 = const()[name = tensor<string, []>("op_8490_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_8490_end_mask_0 = const()[name = tensor<string, []>("op_8490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8490_cast = slice_by_index(begin = var_8490_begin_0, end = var_8490_end_0, end_mask = var_8490_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8490_cast")];
            tensor<int32, [4]> var_8494_begin_0 = const()[name = tensor<string, []>("op_8494_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_8494_end_0 = const()[name = tensor<string, []>("op_8494_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_8494_end_mask_0 = const()[name = tensor<string, []>("op_8494_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8494_cast = slice_by_index(begin = var_8494_begin_0, end = var_8494_end_0, end_mask = var_8494_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8494_cast")];
            tensor<int32, [4]> var_8498_begin_0 = const()[name = tensor<string, []>("op_8498_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_8498_end_0 = const()[name = tensor<string, []>("op_8498_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_8498_end_mask_0 = const()[name = tensor<string, []>("op_8498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8498_cast = slice_by_index(begin = var_8498_begin_0, end = var_8498_end_0, end_mask = var_8498_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8498_cast")];
            tensor<int32, [4]> var_8502_begin_0 = const()[name = tensor<string, []>("op_8502_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_8502_end_0 = const()[name = tensor<string, []>("op_8502_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_8502_end_mask_0 = const()[name = tensor<string, []>("op_8502_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8502_cast = slice_by_index(begin = var_8502_begin_0, end = var_8502_end_0, end_mask = var_8502_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8502_cast")];
            tensor<int32, [4]> var_8506_begin_0 = const()[name = tensor<string, []>("op_8506_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_8506_end_0 = const()[name = tensor<string, []>("op_8506_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_8506_end_mask_0 = const()[name = tensor<string, []>("op_8506_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8506_cast = slice_by_index(begin = var_8506_begin_0, end = var_8506_end_0, end_mask = var_8506_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8506_cast")];
            tensor<int32, [4]> var_8510_begin_0 = const()[name = tensor<string, []>("op_8510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_8510_end_0 = const()[name = tensor<string, []>("op_8510_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_8510_end_mask_0 = const()[name = tensor<string, []>("op_8510_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8510_cast = slice_by_index(begin = var_8510_begin_0, end = var_8510_end_0, end_mask = var_8510_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8510_cast")];
            tensor<int32, [4]> var_8514_begin_0 = const()[name = tensor<string, []>("op_8514_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_8514_end_0 = const()[name = tensor<string, []>("op_8514_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_8514_end_mask_0 = const()[name = tensor<string, []>("op_8514_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_8514_cast = slice_by_index(begin = var_8514_begin_0, end = var_8514_end_0, end_mask = var_8514_end_mask_0, x = transpose_15)[name = tensor<string, []>("op_8514_cast")];
            tensor<int32, [4]> var_8516_begin_0 = const()[name = tensor<string, []>("op_8516_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8516_end_0 = const()[name = tensor<string, []>("op_8516_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_8516_end_mask_0 = const()[name = tensor<string, []>("op_8516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8516_cast = slice_by_index(begin = var_8516_begin_0, end = var_8516_end_0, end_mask = var_8516_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8516_cast")];
            tensor<int32, [4]> var_8520_begin_0 = const()[name = tensor<string, []>("op_8520_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8520_end_0 = const()[name = tensor<string, []>("op_8520_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_8520_end_mask_0 = const()[name = tensor<string, []>("op_8520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8520_cast = slice_by_index(begin = var_8520_begin_0, end = var_8520_end_0, end_mask = var_8520_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8520_cast")];
            tensor<int32, [4]> var_8524_begin_0 = const()[name = tensor<string, []>("op_8524_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8524_end_0 = const()[name = tensor<string, []>("op_8524_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_8524_end_mask_0 = const()[name = tensor<string, []>("op_8524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8524_cast = slice_by_index(begin = var_8524_begin_0, end = var_8524_end_0, end_mask = var_8524_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8524_cast")];
            tensor<int32, [4]> var_8528_begin_0 = const()[name = tensor<string, []>("op_8528_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8528_end_0 = const()[name = tensor<string, []>("op_8528_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_8528_end_mask_0 = const()[name = tensor<string, []>("op_8528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8528_cast = slice_by_index(begin = var_8528_begin_0, end = var_8528_end_0, end_mask = var_8528_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8528_cast")];
            tensor<int32, [4]> var_8532_begin_0 = const()[name = tensor<string, []>("op_8532_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8532_end_0 = const()[name = tensor<string, []>("op_8532_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_8532_end_mask_0 = const()[name = tensor<string, []>("op_8532_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8532_cast = slice_by_index(begin = var_8532_begin_0, end = var_8532_end_0, end_mask = var_8532_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8532_cast")];
            tensor<int32, [4]> var_8536_begin_0 = const()[name = tensor<string, []>("op_8536_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8536_end_0 = const()[name = tensor<string, []>("op_8536_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_8536_end_mask_0 = const()[name = tensor<string, []>("op_8536_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8536_cast = slice_by_index(begin = var_8536_begin_0, end = var_8536_end_0, end_mask = var_8536_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8536_cast")];
            tensor<int32, [4]> var_8540_begin_0 = const()[name = tensor<string, []>("op_8540_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8540_end_0 = const()[name = tensor<string, []>("op_8540_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_8540_end_mask_0 = const()[name = tensor<string, []>("op_8540_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8540_cast = slice_by_index(begin = var_8540_begin_0, end = var_8540_end_0, end_mask = var_8540_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8540_cast")];
            tensor<int32, [4]> var_8544_begin_0 = const()[name = tensor<string, []>("op_8544_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8544_end_0 = const()[name = tensor<string, []>("op_8544_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_8544_end_mask_0 = const()[name = tensor<string, []>("op_8544_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8544_cast = slice_by_index(begin = var_8544_begin_0, end = var_8544_end_0, end_mask = var_8544_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8544_cast")];
            tensor<int32, [4]> var_8548_begin_0 = const()[name = tensor<string, []>("op_8548_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8548_end_0 = const()[name = tensor<string, []>("op_8548_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_8548_end_mask_0 = const()[name = tensor<string, []>("op_8548_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8548_cast = slice_by_index(begin = var_8548_begin_0, end = var_8548_end_0, end_mask = var_8548_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8548_cast")];
            tensor<int32, [4]> var_8552_begin_0 = const()[name = tensor<string, []>("op_8552_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8552_end_0 = const()[name = tensor<string, []>("op_8552_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_8552_end_mask_0 = const()[name = tensor<string, []>("op_8552_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8552_cast = slice_by_index(begin = var_8552_begin_0, end = var_8552_end_0, end_mask = var_8552_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8552_cast")];
            tensor<int32, [4]> var_8556_begin_0 = const()[name = tensor<string, []>("op_8556_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8556_end_0 = const()[name = tensor<string, []>("op_8556_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_8556_end_mask_0 = const()[name = tensor<string, []>("op_8556_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8556_cast = slice_by_index(begin = var_8556_begin_0, end = var_8556_end_0, end_mask = var_8556_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8556_cast")];
            tensor<int32, [4]> var_8560_begin_0 = const()[name = tensor<string, []>("op_8560_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8560_end_0 = const()[name = tensor<string, []>("op_8560_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_8560_end_mask_0 = const()[name = tensor<string, []>("op_8560_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8560_cast = slice_by_index(begin = var_8560_begin_0, end = var_8560_end_0, end_mask = var_8560_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8560_cast")];
            tensor<int32, [4]> var_8564_begin_0 = const()[name = tensor<string, []>("op_8564_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8564_end_0 = const()[name = tensor<string, []>("op_8564_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_8564_end_mask_0 = const()[name = tensor<string, []>("op_8564_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8564_cast = slice_by_index(begin = var_8564_begin_0, end = var_8564_end_0, end_mask = var_8564_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8564_cast")];
            tensor<int32, [4]> var_8568_begin_0 = const()[name = tensor<string, []>("op_8568_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8568_end_0 = const()[name = tensor<string, []>("op_8568_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_8568_end_mask_0 = const()[name = tensor<string, []>("op_8568_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8568_cast = slice_by_index(begin = var_8568_begin_0, end = var_8568_end_0, end_mask = var_8568_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8568_cast")];
            tensor<int32, [4]> var_8572_begin_0 = const()[name = tensor<string, []>("op_8572_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8572_end_0 = const()[name = tensor<string, []>("op_8572_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_8572_end_mask_0 = const()[name = tensor<string, []>("op_8572_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8572_cast = slice_by_index(begin = var_8572_begin_0, end = var_8572_end_0, end_mask = var_8572_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8572_cast")];
            tensor<int32, [4]> var_8576_begin_0 = const()[name = tensor<string, []>("op_8576_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8576_end_0 = const()[name = tensor<string, []>("op_8576_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_8576_end_mask_0 = const()[name = tensor<string, []>("op_8576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8576_cast = slice_by_index(begin = var_8576_begin_0, end = var_8576_end_0, end_mask = var_8576_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8576_cast")];
            tensor<int32, [4]> var_8580_begin_0 = const()[name = tensor<string, []>("op_8580_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8580_end_0 = const()[name = tensor<string, []>("op_8580_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_8580_end_mask_0 = const()[name = tensor<string, []>("op_8580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8580_cast = slice_by_index(begin = var_8580_begin_0, end = var_8580_end_0, end_mask = var_8580_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8580_cast")];
            tensor<int32, [4]> var_8584_begin_0 = const()[name = tensor<string, []>("op_8584_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8584_end_0 = const()[name = tensor<string, []>("op_8584_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_8584_end_mask_0 = const()[name = tensor<string, []>("op_8584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8584_cast = slice_by_index(begin = var_8584_begin_0, end = var_8584_end_0, end_mask = var_8584_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8584_cast")];
            tensor<int32, [4]> var_8588_begin_0 = const()[name = tensor<string, []>("op_8588_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8588_end_0 = const()[name = tensor<string, []>("op_8588_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_8588_end_mask_0 = const()[name = tensor<string, []>("op_8588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8588_cast = slice_by_index(begin = var_8588_begin_0, end = var_8588_end_0, end_mask = var_8588_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8588_cast")];
            tensor<int32, [4]> var_8592_begin_0 = const()[name = tensor<string, []>("op_8592_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8592_end_0 = const()[name = tensor<string, []>("op_8592_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_8592_end_mask_0 = const()[name = tensor<string, []>("op_8592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8592_cast = slice_by_index(begin = var_8592_begin_0, end = var_8592_end_0, end_mask = var_8592_end_mask_0, x = v_33_cast)[name = tensor<string, []>("op_8592_cast")];
            tensor<string, []> var_8596_equation_0 = const()[name = tensor<string, []>("op_8596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8596_cast = einsum(equation = var_8596_equation_0, values = (var_8438_cast, var_8355_cast))[name = tensor<string, []>("op_8596_cast")];
            tensor<fp16, []> var_8597_to_fp16 = const()[name = tensor<string, []>("op_8597_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_321_cast = mul(x = var_8596_cast, y = var_8597_to_fp16)[name = tensor<string, []>("aw_321_cast")];
            tensor<string, []> var_8600_equation_0 = const()[name = tensor<string, []>("op_8600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8600_cast = einsum(equation = var_8600_equation_0, values = (var_8442_cast, var_8359_cast))[name = tensor<string, []>("op_8600_cast")];
            tensor<fp16, []> var_8601_to_fp16 = const()[name = tensor<string, []>("op_8601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_323_cast = mul(x = var_8600_cast, y = var_8601_to_fp16)[name = tensor<string, []>("aw_323_cast")];
            tensor<string, []> var_8604_equation_0 = const()[name = tensor<string, []>("op_8604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8604_cast = einsum(equation = var_8604_equation_0, values = (var_8446_cast, var_8363_cast))[name = tensor<string, []>("op_8604_cast")];
            tensor<fp16, []> var_8605_to_fp16 = const()[name = tensor<string, []>("op_8605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_325_cast = mul(x = var_8604_cast, y = var_8605_to_fp16)[name = tensor<string, []>("aw_325_cast")];
            tensor<string, []> var_8608_equation_0 = const()[name = tensor<string, []>("op_8608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8608_cast = einsum(equation = var_8608_equation_0, values = (var_8450_cast, var_8367_cast))[name = tensor<string, []>("op_8608_cast")];
            tensor<fp16, []> var_8609_to_fp16 = const()[name = tensor<string, []>("op_8609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_327_cast = mul(x = var_8608_cast, y = var_8609_to_fp16)[name = tensor<string, []>("aw_327_cast")];
            tensor<string, []> var_8612_equation_0 = const()[name = tensor<string, []>("op_8612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8612_cast = einsum(equation = var_8612_equation_0, values = (var_8454_cast, var_8371_cast))[name = tensor<string, []>("op_8612_cast")];
            tensor<fp16, []> var_8613_to_fp16 = const()[name = tensor<string, []>("op_8613_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_329_cast = mul(x = var_8612_cast, y = var_8613_to_fp16)[name = tensor<string, []>("aw_329_cast")];
            tensor<string, []> var_8616_equation_0 = const()[name = tensor<string, []>("op_8616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8616_cast = einsum(equation = var_8616_equation_0, values = (var_8458_cast, var_8375_cast))[name = tensor<string, []>("op_8616_cast")];
            tensor<fp16, []> var_8617_to_fp16 = const()[name = tensor<string, []>("op_8617_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_331_cast = mul(x = var_8616_cast, y = var_8617_to_fp16)[name = tensor<string, []>("aw_331_cast")];
            tensor<string, []> var_8620_equation_0 = const()[name = tensor<string, []>("op_8620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8620_cast = einsum(equation = var_8620_equation_0, values = (var_8462_cast, var_8379_cast))[name = tensor<string, []>("op_8620_cast")];
            tensor<fp16, []> var_8621_to_fp16 = const()[name = tensor<string, []>("op_8621_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_333_cast = mul(x = var_8620_cast, y = var_8621_to_fp16)[name = tensor<string, []>("aw_333_cast")];
            tensor<string, []> var_8624_equation_0 = const()[name = tensor<string, []>("op_8624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8624_cast = einsum(equation = var_8624_equation_0, values = (var_8466_cast, var_8383_cast))[name = tensor<string, []>("op_8624_cast")];
            tensor<fp16, []> var_8625_to_fp16 = const()[name = tensor<string, []>("op_8625_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_335_cast = mul(x = var_8624_cast, y = var_8625_to_fp16)[name = tensor<string, []>("aw_335_cast")];
            tensor<string, []> var_8628_equation_0 = const()[name = tensor<string, []>("op_8628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8628_cast = einsum(equation = var_8628_equation_0, values = (var_8470_cast, var_8387_cast))[name = tensor<string, []>("op_8628_cast")];
            tensor<fp16, []> var_8629_to_fp16 = const()[name = tensor<string, []>("op_8629_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_337_cast = mul(x = var_8628_cast, y = var_8629_to_fp16)[name = tensor<string, []>("aw_337_cast")];
            tensor<string, []> var_8632_equation_0 = const()[name = tensor<string, []>("op_8632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8632_cast = einsum(equation = var_8632_equation_0, values = (var_8474_cast, var_8391_cast))[name = tensor<string, []>("op_8632_cast")];
            tensor<fp16, []> var_8633_to_fp16 = const()[name = tensor<string, []>("op_8633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_339_cast = mul(x = var_8632_cast, y = var_8633_to_fp16)[name = tensor<string, []>("aw_339_cast")];
            tensor<string, []> var_8636_equation_0 = const()[name = tensor<string, []>("op_8636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8636_cast = einsum(equation = var_8636_equation_0, values = (var_8478_cast, var_8395_cast))[name = tensor<string, []>("op_8636_cast")];
            tensor<fp16, []> var_8637_to_fp16 = const()[name = tensor<string, []>("op_8637_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_341_cast = mul(x = var_8636_cast, y = var_8637_to_fp16)[name = tensor<string, []>("aw_341_cast")];
            tensor<string, []> var_8640_equation_0 = const()[name = tensor<string, []>("op_8640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8640_cast = einsum(equation = var_8640_equation_0, values = (var_8482_cast, var_8399_cast))[name = tensor<string, []>("op_8640_cast")];
            tensor<fp16, []> var_8641_to_fp16 = const()[name = tensor<string, []>("op_8641_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_343_cast = mul(x = var_8640_cast, y = var_8641_to_fp16)[name = tensor<string, []>("aw_343_cast")];
            tensor<string, []> var_8644_equation_0 = const()[name = tensor<string, []>("op_8644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8644_cast = einsum(equation = var_8644_equation_0, values = (var_8486_cast, var_8403_cast))[name = tensor<string, []>("op_8644_cast")];
            tensor<fp16, []> var_8645_to_fp16 = const()[name = tensor<string, []>("op_8645_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_345_cast = mul(x = var_8644_cast, y = var_8645_to_fp16)[name = tensor<string, []>("aw_345_cast")];
            tensor<string, []> var_8648_equation_0 = const()[name = tensor<string, []>("op_8648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8648_cast = einsum(equation = var_8648_equation_0, values = (var_8490_cast, var_8407_cast))[name = tensor<string, []>("op_8648_cast")];
            tensor<fp16, []> var_8649_to_fp16 = const()[name = tensor<string, []>("op_8649_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_347_cast = mul(x = var_8648_cast, y = var_8649_to_fp16)[name = tensor<string, []>("aw_347_cast")];
            tensor<string, []> var_8652_equation_0 = const()[name = tensor<string, []>("op_8652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8652_cast = einsum(equation = var_8652_equation_0, values = (var_8494_cast, var_8411_cast))[name = tensor<string, []>("op_8652_cast")];
            tensor<fp16, []> var_8653_to_fp16 = const()[name = tensor<string, []>("op_8653_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_349_cast = mul(x = var_8652_cast, y = var_8653_to_fp16)[name = tensor<string, []>("aw_349_cast")];
            tensor<string, []> var_8656_equation_0 = const()[name = tensor<string, []>("op_8656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8656_cast = einsum(equation = var_8656_equation_0, values = (var_8498_cast, var_8415_cast))[name = tensor<string, []>("op_8656_cast")];
            tensor<fp16, []> var_8657_to_fp16 = const()[name = tensor<string, []>("op_8657_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_351_cast = mul(x = var_8656_cast, y = var_8657_to_fp16)[name = tensor<string, []>("aw_351_cast")];
            tensor<string, []> var_8660_equation_0 = const()[name = tensor<string, []>("op_8660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8660_cast = einsum(equation = var_8660_equation_0, values = (var_8502_cast, var_8419_cast))[name = tensor<string, []>("op_8660_cast")];
            tensor<fp16, []> var_8661_to_fp16 = const()[name = tensor<string, []>("op_8661_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_353_cast = mul(x = var_8660_cast, y = var_8661_to_fp16)[name = tensor<string, []>("aw_353_cast")];
            tensor<string, []> var_8664_equation_0 = const()[name = tensor<string, []>("op_8664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8664_cast = einsum(equation = var_8664_equation_0, values = (var_8506_cast, var_8423_cast))[name = tensor<string, []>("op_8664_cast")];
            tensor<fp16, []> var_8665_to_fp16 = const()[name = tensor<string, []>("op_8665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_355_cast = mul(x = var_8664_cast, y = var_8665_to_fp16)[name = tensor<string, []>("aw_355_cast")];
            tensor<string, []> var_8668_equation_0 = const()[name = tensor<string, []>("op_8668_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8668_cast = einsum(equation = var_8668_equation_0, values = (var_8510_cast, var_8427_cast))[name = tensor<string, []>("op_8668_cast")];
            tensor<fp16, []> var_8669_to_fp16 = const()[name = tensor<string, []>("op_8669_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_357_cast = mul(x = var_8668_cast, y = var_8669_to_fp16)[name = tensor<string, []>("aw_357_cast")];
            tensor<string, []> var_8672_equation_0 = const()[name = tensor<string, []>("op_8672_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_8672_cast = einsum(equation = var_8672_equation_0, values = (var_8514_cast, var_8431_cast))[name = tensor<string, []>("op_8672_cast")];
            tensor<fp16, []> var_8673_to_fp16 = const()[name = tensor<string, []>("op_8673_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_359_cast = mul(x = var_8672_cast, y = var_8673_to_fp16)[name = tensor<string, []>("aw_359_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8675_cast = softmax(axis = var_7228, x = aw_321_cast)[name = tensor<string, []>("op_8675_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8676_cast = softmax(axis = var_7228, x = aw_323_cast)[name = tensor<string, []>("op_8676_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8677_cast = softmax(axis = var_7228, x = aw_325_cast)[name = tensor<string, []>("op_8677_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8678_cast = softmax(axis = var_7228, x = aw_327_cast)[name = tensor<string, []>("op_8678_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8679_cast = softmax(axis = var_7228, x = aw_329_cast)[name = tensor<string, []>("op_8679_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8680_cast = softmax(axis = var_7228, x = aw_331_cast)[name = tensor<string, []>("op_8680_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8681_cast = softmax(axis = var_7228, x = aw_333_cast)[name = tensor<string, []>("op_8681_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8682_cast = softmax(axis = var_7228, x = aw_335_cast)[name = tensor<string, []>("op_8682_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8683_cast = softmax(axis = var_7228, x = aw_337_cast)[name = tensor<string, []>("op_8683_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8684_cast = softmax(axis = var_7228, x = aw_339_cast)[name = tensor<string, []>("op_8684_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8685_cast = softmax(axis = var_7228, x = aw_341_cast)[name = tensor<string, []>("op_8685_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8686_cast = softmax(axis = var_7228, x = aw_343_cast)[name = tensor<string, []>("op_8686_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8687_cast = softmax(axis = var_7228, x = aw_345_cast)[name = tensor<string, []>("op_8687_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8688_cast = softmax(axis = var_7228, x = aw_347_cast)[name = tensor<string, []>("op_8688_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8689_cast = softmax(axis = var_7228, x = aw_349_cast)[name = tensor<string, []>("op_8689_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8690_cast = softmax(axis = var_7228, x = aw_351_cast)[name = tensor<string, []>("op_8690_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8691_cast = softmax(axis = var_7228, x = aw_353_cast)[name = tensor<string, []>("op_8691_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8692_cast = softmax(axis = var_7228, x = aw_355_cast)[name = tensor<string, []>("op_8692_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8693_cast = softmax(axis = var_7228, x = aw_357_cast)[name = tensor<string, []>("op_8693_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_8694_cast = softmax(axis = var_7228, x = aw_359_cast)[name = tensor<string, []>("op_8694_cast")];
            tensor<string, []> var_8696_equation_0 = const()[name = tensor<string, []>("op_8696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8696_cast = einsum(equation = var_8696_equation_0, values = (var_8516_cast, var_8675_cast))[name = tensor<string, []>("op_8696_cast")];
            tensor<string, []> var_8698_equation_0 = const()[name = tensor<string, []>("op_8698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8698_cast = einsum(equation = var_8698_equation_0, values = (var_8520_cast, var_8676_cast))[name = tensor<string, []>("op_8698_cast")];
            tensor<string, []> var_8700_equation_0 = const()[name = tensor<string, []>("op_8700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8700_cast = einsum(equation = var_8700_equation_0, values = (var_8524_cast, var_8677_cast))[name = tensor<string, []>("op_8700_cast")];
            tensor<string, []> var_8702_equation_0 = const()[name = tensor<string, []>("op_8702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8702_cast = einsum(equation = var_8702_equation_0, values = (var_8528_cast, var_8678_cast))[name = tensor<string, []>("op_8702_cast")];
            tensor<string, []> var_8704_equation_0 = const()[name = tensor<string, []>("op_8704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8704_cast = einsum(equation = var_8704_equation_0, values = (var_8532_cast, var_8679_cast))[name = tensor<string, []>("op_8704_cast")];
            tensor<string, []> var_8706_equation_0 = const()[name = tensor<string, []>("op_8706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8706_cast = einsum(equation = var_8706_equation_0, values = (var_8536_cast, var_8680_cast))[name = tensor<string, []>("op_8706_cast")];
            tensor<string, []> var_8708_equation_0 = const()[name = tensor<string, []>("op_8708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8708_cast = einsum(equation = var_8708_equation_0, values = (var_8540_cast, var_8681_cast))[name = tensor<string, []>("op_8708_cast")];
            tensor<string, []> var_8710_equation_0 = const()[name = tensor<string, []>("op_8710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8710_cast = einsum(equation = var_8710_equation_0, values = (var_8544_cast, var_8682_cast))[name = tensor<string, []>("op_8710_cast")];
            tensor<string, []> var_8712_equation_0 = const()[name = tensor<string, []>("op_8712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8712_cast = einsum(equation = var_8712_equation_0, values = (var_8548_cast, var_8683_cast))[name = tensor<string, []>("op_8712_cast")];
            tensor<string, []> var_8714_equation_0 = const()[name = tensor<string, []>("op_8714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8714_cast = einsum(equation = var_8714_equation_0, values = (var_8552_cast, var_8684_cast))[name = tensor<string, []>("op_8714_cast")];
            tensor<string, []> var_8716_equation_0 = const()[name = tensor<string, []>("op_8716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8716_cast = einsum(equation = var_8716_equation_0, values = (var_8556_cast, var_8685_cast))[name = tensor<string, []>("op_8716_cast")];
            tensor<string, []> var_8718_equation_0 = const()[name = tensor<string, []>("op_8718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8718_cast = einsum(equation = var_8718_equation_0, values = (var_8560_cast, var_8686_cast))[name = tensor<string, []>("op_8718_cast")];
            tensor<string, []> var_8720_equation_0 = const()[name = tensor<string, []>("op_8720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8720_cast = einsum(equation = var_8720_equation_0, values = (var_8564_cast, var_8687_cast))[name = tensor<string, []>("op_8720_cast")];
            tensor<string, []> var_8722_equation_0 = const()[name = tensor<string, []>("op_8722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8722_cast = einsum(equation = var_8722_equation_0, values = (var_8568_cast, var_8688_cast))[name = tensor<string, []>("op_8722_cast")];
            tensor<string, []> var_8724_equation_0 = const()[name = tensor<string, []>("op_8724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8724_cast = einsum(equation = var_8724_equation_0, values = (var_8572_cast, var_8689_cast))[name = tensor<string, []>("op_8724_cast")];
            tensor<string, []> var_8726_equation_0 = const()[name = tensor<string, []>("op_8726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8726_cast = einsum(equation = var_8726_equation_0, values = (var_8576_cast, var_8690_cast))[name = tensor<string, []>("op_8726_cast")];
            tensor<string, []> var_8728_equation_0 = const()[name = tensor<string, []>("op_8728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8728_cast = einsum(equation = var_8728_equation_0, values = (var_8580_cast, var_8691_cast))[name = tensor<string, []>("op_8728_cast")];
            tensor<string, []> var_8730_equation_0 = const()[name = tensor<string, []>("op_8730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8730_cast = einsum(equation = var_8730_equation_0, values = (var_8584_cast, var_8692_cast))[name = tensor<string, []>("op_8730_cast")];
            tensor<string, []> var_8732_equation_0 = const()[name = tensor<string, []>("op_8732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8732_cast = einsum(equation = var_8732_equation_0, values = (var_8588_cast, var_8693_cast))[name = tensor<string, []>("op_8732_cast")];
            tensor<string, []> var_8734_equation_0 = const()[name = tensor<string, []>("op_8734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_8734_cast = einsum(equation = var_8734_equation_0, values = (var_8592_cast, var_8694_cast))[name = tensor<string, []>("op_8734_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_7228, interleave = input_327_interleave_0, values = (var_8696_cast, var_8698_cast, var_8700_cast, var_8702_cast, var_8704_cast, var_8706_cast, var_8708_cast, var_8710_cast, var_8712_cast, var_8714_cast, var_8716_cast, var_8718_cast, var_8720_cast, var_8722_cast, var_8724_cast, var_8726_cast, var_8728_cast, var_8730_cast, var_8732_cast, var_8734_cast))[name = tensor<string, []>("input_327_cast")];
            tensor<int32, [2]> var_8740 = const()[name = tensor<string, []>("op_8740"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8742 = const()[name = tensor<string, []>("op_8742"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_8744_pad_type_0 = const()[name = tensor<string, []>("op_8744_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_8744_pad_0 = const()[name = tensor<string, []>("op_8744_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163956160))), lut = tensor<fp16, [4]>([-0x1.644p-4, -0x1.aacp-6, 0x1.a8cp-6, 0x1.63cp-4]), 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, []>(164365824)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8744_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_8742, groups = var_7228, pad = var_8744_pad_0, pad_type = var_8744_pad_type_0, strides = var_8740, 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_8744_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_51_cast = add(x = var_8744_cast, y = inputs_49_cast)[name = tensor<string, []>("inputs_51_cast")];
            tensor<int32, [1]> var_8748 = const()[name = tensor<string, []>("op_8748"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_51_cast = reduce_mean(axes = var_8748, keep_dims = var_7223, 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_8752 = const()[name = tensor<string, []>("op_8752"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_8753_cast = reduce_mean(axes = var_8752, keep_dims = var_7223, x = zero_mean_sq_51_cast)[name = tensor<string, []>("op_8753_cast")];
            tensor<fp16, []> var_8754_to_fp16 = const()[name = tensor<string, []>("op_8754_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_8755_cast = add(x = var_8753_cast, y = var_8754_to_fp16)[name = tensor<string, []>("op_8755_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_8755_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_8759_to_fp16 = const()[name = tensor<string, []>("op_8759_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164368448)))];
            tensor<fp16, [2, 1280, 1, 256]> var_8760_cast = add(x = out_51_cast, y = var_8759_to_fp16)[name = tensor<string, []>("op_8760_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_8762_to_fp16 = const()[name = tensor<string, []>("op_8762_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164371072)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_189_cast = mul(x = var_8760_cast, y = var_8762_to_fp16)[name = tensor<string, []>("hidden_states_189_cast")];
            tensor<int32, [2]> var_8769 = const()[name = tensor<string, []>("op_8769"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8771 = const()[name = tensor<string, []>("op_8771"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164373696))), lut = tensor<fp16, [4]>([-0x1.3cp-4, -0x1.7c4p-6, 0x1.72cp-6, 0x1.3ap-4]), 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_8771, groups = var_7228, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = var_8769, 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_8775 = const()[name = tensor<string, []>("op_8775"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8777 = const()[name = tensor<string, []>("op_8777"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_69_pad_type_0 = const()[name = tensor<string, []>("k_69_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_69_pad_0 = const()[name = tensor<string, []>("k_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164783360))), lut = tensor<fp16, [4]>([-0x1.484p-4, -0x1.834p-6, 0x1.87p-6, 0x1.494p-4]), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_69_cast = conv(dilations = var_8777, groups = var_7228, pad = k_69_pad_0, pad_type = k_69_pad_type_0, strides = var_8775, 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_8781 = const()[name = tensor<string, []>("op_8781"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_8783 = const()[name = tensor<string, []>("op_8783"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_35_pad_type_0 = const()[name = tensor<string, []>("v_35_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_35_pad_0 = const()[name = tensor<string, []>("v_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165111104))), lut = tensor<fp16, [4]>([-0x1.fecp-5, -0x1.2ap-6, 0x1.2ccp-6, 0x1.008p-4]), name = tensor<string, []>("up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_35_cast = conv(dilations = var_8783, groups = var_7228, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = var_8781, 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_8787_begin_0 = const()[name = tensor<string, []>("op_8787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8787_end_0 = const()[name = tensor<string, []>("op_8787_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_8787_end_mask_0 = const()[name = tensor<string, []>("op_8787_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8787_cast = slice_by_index(begin = var_8787_begin_0, end = var_8787_end_0, end_mask = var_8787_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8787_cast")];
            tensor<int32, [4]> var_8791_begin_0 = const()[name = tensor<string, []>("op_8791_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8791_end_0 = const()[name = tensor<string, []>("op_8791_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_8791_end_mask_0 = const()[name = tensor<string, []>("op_8791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8791_cast = slice_by_index(begin = var_8791_begin_0, end = var_8791_end_0, end_mask = var_8791_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8791_cast")];
            tensor<int32, [4]> var_8795_begin_0 = const()[name = tensor<string, []>("op_8795_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8795_end_0 = const()[name = tensor<string, []>("op_8795_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_8795_end_mask_0 = const()[name = tensor<string, []>("op_8795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8795_cast = slice_by_index(begin = var_8795_begin_0, end = var_8795_end_0, end_mask = var_8795_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8795_cast")];
            tensor<int32, [4]> var_8799_begin_0 = const()[name = tensor<string, []>("op_8799_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8799_end_0 = const()[name = tensor<string, []>("op_8799_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_8799_end_mask_0 = const()[name = tensor<string, []>("op_8799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8799_cast = slice_by_index(begin = var_8799_begin_0, end = var_8799_end_0, end_mask = var_8799_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8799_cast")];
            tensor<int32, [4]> var_8803_begin_0 = const()[name = tensor<string, []>("op_8803_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8803_end_0 = const()[name = tensor<string, []>("op_8803_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_8803_end_mask_0 = const()[name = tensor<string, []>("op_8803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8803_cast = slice_by_index(begin = var_8803_begin_0, end = var_8803_end_0, end_mask = var_8803_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8803_cast")];
            tensor<int32, [4]> var_8807_begin_0 = const()[name = tensor<string, []>("op_8807_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8807_end_0 = const()[name = tensor<string, []>("op_8807_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_8807_end_mask_0 = const()[name = tensor<string, []>("op_8807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8807_cast = slice_by_index(begin = var_8807_begin_0, end = var_8807_end_0, end_mask = var_8807_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8807_cast")];
            tensor<int32, [4]> var_8811_begin_0 = const()[name = tensor<string, []>("op_8811_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8811_end_0 = const()[name = tensor<string, []>("op_8811_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_8811_end_mask_0 = const()[name = tensor<string, []>("op_8811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8811_cast = slice_by_index(begin = var_8811_begin_0, end = var_8811_end_0, end_mask = var_8811_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8811_cast")];
            tensor<int32, [4]> var_8815_begin_0 = const()[name = tensor<string, []>("op_8815_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8815_end_0 = const()[name = tensor<string, []>("op_8815_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_8815_end_mask_0 = const()[name = tensor<string, []>("op_8815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8815_cast = slice_by_index(begin = var_8815_begin_0, end = var_8815_end_0, end_mask = var_8815_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8815_cast")];
            tensor<int32, [4]> var_8819_begin_0 = const()[name = tensor<string, []>("op_8819_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8819_end_0 = const()[name = tensor<string, []>("op_8819_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_8819_end_mask_0 = const()[name = tensor<string, []>("op_8819_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8819_cast = slice_by_index(begin = var_8819_begin_0, end = var_8819_end_0, end_mask = var_8819_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8819_cast")];
            tensor<int32, [4]> var_8823_begin_0 = const()[name = tensor<string, []>("op_8823_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8823_end_0 = const()[name = tensor<string, []>("op_8823_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_8823_end_mask_0 = const()[name = tensor<string, []>("op_8823_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8823_cast = slice_by_index(begin = var_8823_begin_0, end = var_8823_end_0, end_mask = var_8823_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8823_cast")];
            tensor<int32, [4]> var_8827_begin_0 = const()[name = tensor<string, []>("op_8827_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8827_end_0 = const()[name = tensor<string, []>("op_8827_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_8827_end_mask_0 = const()[name = tensor<string, []>("op_8827_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8827_cast = slice_by_index(begin = var_8827_begin_0, end = var_8827_end_0, end_mask = var_8827_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8827_cast")];
            tensor<int32, [4]> var_8831_begin_0 = const()[name = tensor<string, []>("op_8831_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8831_end_0 = const()[name = tensor<string, []>("op_8831_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_8831_end_mask_0 = const()[name = tensor<string, []>("op_8831_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8831_cast = slice_by_index(begin = var_8831_begin_0, end = var_8831_end_0, end_mask = var_8831_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8831_cast")];
            tensor<int32, [4]> var_8835_begin_0 = const()[name = tensor<string, []>("op_8835_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8835_end_0 = const()[name = tensor<string, []>("op_8835_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_8835_end_mask_0 = const()[name = tensor<string, []>("op_8835_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8835_cast = slice_by_index(begin = var_8835_begin_0, end = var_8835_end_0, end_mask = var_8835_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8835_cast")];
            tensor<int32, [4]> var_8839_begin_0 = const()[name = tensor<string, []>("op_8839_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_8839_end_0 = const()[name = tensor<string, []>("op_8839_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_8839_end_mask_0 = const()[name = tensor<string, []>("op_8839_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8839_cast = slice_by_index(begin = var_8839_begin_0, end = var_8839_end_0, end_mask = var_8839_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8839_cast")];
            tensor<int32, [4]> var_8843_begin_0 = const()[name = tensor<string, []>("op_8843_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_8843_end_0 = const()[name = tensor<string, []>("op_8843_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_8843_end_mask_0 = const()[name = tensor<string, []>("op_8843_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8843_cast = slice_by_index(begin = var_8843_begin_0, end = var_8843_end_0, end_mask = var_8843_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8843_cast")];
            tensor<int32, [4]> var_8847_begin_0 = const()[name = tensor<string, []>("op_8847_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_8847_end_0 = const()[name = tensor<string, []>("op_8847_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_8847_end_mask_0 = const()[name = tensor<string, []>("op_8847_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8847_cast = slice_by_index(begin = var_8847_begin_0, end = var_8847_end_0, end_mask = var_8847_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8847_cast")];
            tensor<int32, [4]> var_8851_begin_0 = const()[name = tensor<string, []>("op_8851_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_8851_end_0 = const()[name = tensor<string, []>("op_8851_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_8851_end_mask_0 = const()[name = tensor<string, []>("op_8851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8851_cast = slice_by_index(begin = var_8851_begin_0, end = var_8851_end_0, end_mask = var_8851_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8851_cast")];
            tensor<int32, [4]> var_8855_begin_0 = const()[name = tensor<string, []>("op_8855_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_8855_end_0 = const()[name = tensor<string, []>("op_8855_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_8855_end_mask_0 = const()[name = tensor<string, []>("op_8855_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8855_cast = slice_by_index(begin = var_8855_begin_0, end = var_8855_end_0, end_mask = var_8855_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8855_cast")];
            tensor<int32, [4]> var_8859_begin_0 = const()[name = tensor<string, []>("op_8859_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_8859_end_0 = const()[name = tensor<string, []>("op_8859_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_8859_end_mask_0 = const()[name = tensor<string, []>("op_8859_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8859_cast = slice_by_index(begin = var_8859_begin_0, end = var_8859_end_0, end_mask = var_8859_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8859_cast")];
            tensor<int32, [4]> var_8863_begin_0 = const()[name = tensor<string, []>("op_8863_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_8863_end_0 = const()[name = tensor<string, []>("op_8863_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_8863_end_mask_0 = const()[name = tensor<string, []>("op_8863_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_8863_cast = slice_by_index(begin = var_8863_begin_0, end = var_8863_end_0, end_mask = var_8863_end_mask_0, x = q_35_cast)[name = tensor<string, []>("op_8863_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_8870_begin_0 = const()[name = tensor<string, []>("op_8870_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8870_end_0 = const()[name = tensor<string, []>("op_8870_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_8870_end_mask_0 = const()[name = tensor<string, []>("op_8870_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, 64]> var_8870_cast = slice_by_index(begin = var_8870_begin_0, end = var_8870_end_0, end_mask = var_8870_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8870_cast")];
            tensor<int32, [4]> var_8874_begin_0 = const()[name = tensor<string, []>("op_8874_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_8874_end_0 = const()[name = tensor<string, []>("op_8874_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_8874_end_mask_0 = const()[name = tensor<string, []>("op_8874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8874_cast = slice_by_index(begin = var_8874_begin_0, end = var_8874_end_0, end_mask = var_8874_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8874_cast")];
            tensor<int32, [4]> var_8878_begin_0 = const()[name = tensor<string, []>("op_8878_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_8878_end_0 = const()[name = tensor<string, []>("op_8878_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_8878_end_mask_0 = const()[name = tensor<string, []>("op_8878_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8878_cast = slice_by_index(begin = var_8878_begin_0, end = var_8878_end_0, end_mask = var_8878_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8878_cast")];
            tensor<int32, [4]> var_8882_begin_0 = const()[name = tensor<string, []>("op_8882_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_8882_end_0 = const()[name = tensor<string, []>("op_8882_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_8882_end_mask_0 = const()[name = tensor<string, []>("op_8882_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8882_cast = slice_by_index(begin = var_8882_begin_0, end = var_8882_end_0, end_mask = var_8882_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8882_cast")];
            tensor<int32, [4]> var_8886_begin_0 = const()[name = tensor<string, []>("op_8886_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_8886_end_0 = const()[name = tensor<string, []>("op_8886_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_8886_end_mask_0 = const()[name = tensor<string, []>("op_8886_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8886_cast = slice_by_index(begin = var_8886_begin_0, end = var_8886_end_0, end_mask = var_8886_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8886_cast")];
            tensor<int32, [4]> var_8890_begin_0 = const()[name = tensor<string, []>("op_8890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_8890_end_0 = const()[name = tensor<string, []>("op_8890_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_8890_end_mask_0 = const()[name = tensor<string, []>("op_8890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8890_cast = slice_by_index(begin = var_8890_begin_0, end = var_8890_end_0, end_mask = var_8890_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8890_cast")];
            tensor<int32, [4]> var_8894_begin_0 = const()[name = tensor<string, []>("op_8894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_8894_end_0 = const()[name = tensor<string, []>("op_8894_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_8894_end_mask_0 = const()[name = tensor<string, []>("op_8894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8894_cast = slice_by_index(begin = var_8894_begin_0, end = var_8894_end_0, end_mask = var_8894_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8894_cast")];
            tensor<int32, [4]> var_8898_begin_0 = const()[name = tensor<string, []>("op_8898_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_8898_end_0 = const()[name = tensor<string, []>("op_8898_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_8898_end_mask_0 = const()[name = tensor<string, []>("op_8898_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8898_cast = slice_by_index(begin = var_8898_begin_0, end = var_8898_end_0, end_mask = var_8898_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8898_cast")];
            tensor<int32, [4]> var_8902_begin_0 = const()[name = tensor<string, []>("op_8902_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_8902_end_0 = const()[name = tensor<string, []>("op_8902_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_8902_end_mask_0 = const()[name = tensor<string, []>("op_8902_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8902_cast = slice_by_index(begin = var_8902_begin_0, end = var_8902_end_0, end_mask = var_8902_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8902_cast")];
            tensor<int32, [4]> var_8906_begin_0 = const()[name = tensor<string, []>("op_8906_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_8906_end_0 = const()[name = tensor<string, []>("op_8906_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_8906_end_mask_0 = const()[name = tensor<string, []>("op_8906_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8906_cast = slice_by_index(begin = var_8906_begin_0, end = var_8906_end_0, end_mask = var_8906_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8906_cast")];
            tensor<int32, [4]> var_8910_begin_0 = const()[name = tensor<string, []>("op_8910_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_8910_end_0 = const()[name = tensor<string, []>("op_8910_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_8910_end_mask_0 = const()[name = tensor<string, []>("op_8910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8910_cast = slice_by_index(begin = var_8910_begin_0, end = var_8910_end_0, end_mask = var_8910_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8910_cast")];
            tensor<int32, [4]> var_8914_begin_0 = const()[name = tensor<string, []>("op_8914_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_8914_end_0 = const()[name = tensor<string, []>("op_8914_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_8914_end_mask_0 = const()[name = tensor<string, []>("op_8914_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8914_cast = slice_by_index(begin = var_8914_begin_0, end = var_8914_end_0, end_mask = var_8914_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8914_cast")];
            tensor<int32, [4]> var_8918_begin_0 = const()[name = tensor<string, []>("op_8918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_8918_end_0 = const()[name = tensor<string, []>("op_8918_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_8918_end_mask_0 = const()[name = tensor<string, []>("op_8918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8918_cast = slice_by_index(begin = var_8918_begin_0, end = var_8918_end_0, end_mask = var_8918_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8918_cast")];
            tensor<int32, [4]> var_8922_begin_0 = const()[name = tensor<string, []>("op_8922_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_8922_end_0 = const()[name = tensor<string, []>("op_8922_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_8922_end_mask_0 = const()[name = tensor<string, []>("op_8922_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8922_cast = slice_by_index(begin = var_8922_begin_0, end = var_8922_end_0, end_mask = var_8922_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8922_cast")];
            tensor<int32, [4]> var_8926_begin_0 = const()[name = tensor<string, []>("op_8926_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_8926_end_0 = const()[name = tensor<string, []>("op_8926_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_8926_end_mask_0 = const()[name = tensor<string, []>("op_8926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8926_cast = slice_by_index(begin = var_8926_begin_0, end = var_8926_end_0, end_mask = var_8926_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8926_cast")];
            tensor<int32, [4]> var_8930_begin_0 = const()[name = tensor<string, []>("op_8930_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_8930_end_0 = const()[name = tensor<string, []>("op_8930_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_8930_end_mask_0 = const()[name = tensor<string, []>("op_8930_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8930_cast = slice_by_index(begin = var_8930_begin_0, end = var_8930_end_0, end_mask = var_8930_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8930_cast")];
            tensor<int32, [4]> var_8934_begin_0 = const()[name = tensor<string, []>("op_8934_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_8934_end_0 = const()[name = tensor<string, []>("op_8934_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_8934_end_mask_0 = const()[name = tensor<string, []>("op_8934_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8934_cast = slice_by_index(begin = var_8934_begin_0, end = var_8934_end_0, end_mask = var_8934_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8934_cast")];
            tensor<int32, [4]> var_8938_begin_0 = const()[name = tensor<string, []>("op_8938_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_8938_end_0 = const()[name = tensor<string, []>("op_8938_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_8938_end_mask_0 = const()[name = tensor<string, []>("op_8938_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8938_cast = slice_by_index(begin = var_8938_begin_0, end = var_8938_end_0, end_mask = var_8938_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8938_cast")];
            tensor<int32, [4]> var_8942_begin_0 = const()[name = tensor<string, []>("op_8942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_8942_end_0 = const()[name = tensor<string, []>("op_8942_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_8942_end_mask_0 = const()[name = tensor<string, []>("op_8942_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8942_cast = slice_by_index(begin = var_8942_begin_0, end = var_8942_end_0, end_mask = var_8942_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8942_cast")];
            tensor<int32, [4]> var_8946_begin_0 = const()[name = tensor<string, []>("op_8946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_8946_end_0 = const()[name = tensor<string, []>("op_8946_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_8946_end_mask_0 = const()[name = tensor<string, []>("op_8946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_8946_cast = slice_by_index(begin = var_8946_begin_0, end = var_8946_end_0, end_mask = var_8946_end_mask_0, x = transpose_14)[name = tensor<string, []>("op_8946_cast")];
            tensor<int32, [4]> var_8948_begin_0 = const()[name = tensor<string, []>("op_8948_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_8948_end_0 = const()[name = tensor<string, []>("op_8948_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_8948_end_mask_0 = const()[name = tensor<string, []>("op_8948_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8948_cast = slice_by_index(begin = var_8948_begin_0, end = var_8948_end_0, end_mask = var_8948_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8948_cast")];
            tensor<int32, [4]> var_8952_begin_0 = const()[name = tensor<string, []>("op_8952_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_8952_end_0 = const()[name = tensor<string, []>("op_8952_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_8952_end_mask_0 = const()[name = tensor<string, []>("op_8952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8952_cast = slice_by_index(begin = var_8952_begin_0, end = var_8952_end_0, end_mask = var_8952_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8952_cast")];
            tensor<int32, [4]> var_8956_begin_0 = const()[name = tensor<string, []>("op_8956_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_8956_end_0 = const()[name = tensor<string, []>("op_8956_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_8956_end_mask_0 = const()[name = tensor<string, []>("op_8956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8956_cast = slice_by_index(begin = var_8956_begin_0, end = var_8956_end_0, end_mask = var_8956_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8956_cast")];
            tensor<int32, [4]> var_8960_begin_0 = const()[name = tensor<string, []>("op_8960_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_8960_end_0 = const()[name = tensor<string, []>("op_8960_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_8960_end_mask_0 = const()[name = tensor<string, []>("op_8960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8960_cast = slice_by_index(begin = var_8960_begin_0, end = var_8960_end_0, end_mask = var_8960_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8960_cast")];
            tensor<int32, [4]> var_8964_begin_0 = const()[name = tensor<string, []>("op_8964_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_8964_end_0 = const()[name = tensor<string, []>("op_8964_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_8964_end_mask_0 = const()[name = tensor<string, []>("op_8964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8964_cast = slice_by_index(begin = var_8964_begin_0, end = var_8964_end_0, end_mask = var_8964_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8964_cast")];
            tensor<int32, [4]> var_8968_begin_0 = const()[name = tensor<string, []>("op_8968_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_8968_end_0 = const()[name = tensor<string, []>("op_8968_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_8968_end_mask_0 = const()[name = tensor<string, []>("op_8968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8968_cast = slice_by_index(begin = var_8968_begin_0, end = var_8968_end_0, end_mask = var_8968_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8968_cast")];
            tensor<int32, [4]> var_8972_begin_0 = const()[name = tensor<string, []>("op_8972_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_8972_end_0 = const()[name = tensor<string, []>("op_8972_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_8972_end_mask_0 = const()[name = tensor<string, []>("op_8972_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8972_cast = slice_by_index(begin = var_8972_begin_0, end = var_8972_end_0, end_mask = var_8972_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8972_cast")];
            tensor<int32, [4]> var_8976_begin_0 = const()[name = tensor<string, []>("op_8976_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_8976_end_0 = const()[name = tensor<string, []>("op_8976_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_8976_end_mask_0 = const()[name = tensor<string, []>("op_8976_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8976_cast = slice_by_index(begin = var_8976_begin_0, end = var_8976_end_0, end_mask = var_8976_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8976_cast")];
            tensor<int32, [4]> var_8980_begin_0 = const()[name = tensor<string, []>("op_8980_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_8980_end_0 = const()[name = tensor<string, []>("op_8980_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_8980_end_mask_0 = const()[name = tensor<string, []>("op_8980_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8980_cast = slice_by_index(begin = var_8980_begin_0, end = var_8980_end_0, end_mask = var_8980_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8980_cast")];
            tensor<int32, [4]> var_8984_begin_0 = const()[name = tensor<string, []>("op_8984_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_8984_end_0 = const()[name = tensor<string, []>("op_8984_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_8984_end_mask_0 = const()[name = tensor<string, []>("op_8984_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8984_cast = slice_by_index(begin = var_8984_begin_0, end = var_8984_end_0, end_mask = var_8984_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8984_cast")];
            tensor<int32, [4]> var_8988_begin_0 = const()[name = tensor<string, []>("op_8988_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_8988_end_0 = const()[name = tensor<string, []>("op_8988_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_8988_end_mask_0 = const()[name = tensor<string, []>("op_8988_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8988_cast = slice_by_index(begin = var_8988_begin_0, end = var_8988_end_0, end_mask = var_8988_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8988_cast")];
            tensor<int32, [4]> var_8992_begin_0 = const()[name = tensor<string, []>("op_8992_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_8992_end_0 = const()[name = tensor<string, []>("op_8992_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_8992_end_mask_0 = const()[name = tensor<string, []>("op_8992_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8992_cast = slice_by_index(begin = var_8992_begin_0, end = var_8992_end_0, end_mask = var_8992_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8992_cast")];
            tensor<int32, [4]> var_8996_begin_0 = const()[name = tensor<string, []>("op_8996_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_8996_end_0 = const()[name = tensor<string, []>("op_8996_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_8996_end_mask_0 = const()[name = tensor<string, []>("op_8996_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_8996_cast = slice_by_index(begin = var_8996_begin_0, end = var_8996_end_0, end_mask = var_8996_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_8996_cast")];
            tensor<int32, [4]> var_9000_begin_0 = const()[name = tensor<string, []>("op_9000_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9000_end_0 = const()[name = tensor<string, []>("op_9000_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_9000_end_mask_0 = const()[name = tensor<string, []>("op_9000_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9000_cast = slice_by_index(begin = var_9000_begin_0, end = var_9000_end_0, end_mask = var_9000_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9000_cast")];
            tensor<int32, [4]> var_9004_begin_0 = const()[name = tensor<string, []>("op_9004_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9004_end_0 = const()[name = tensor<string, []>("op_9004_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_9004_end_mask_0 = const()[name = tensor<string, []>("op_9004_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9004_cast = slice_by_index(begin = var_9004_begin_0, end = var_9004_end_0, end_mask = var_9004_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9004_cast")];
            tensor<int32, [4]> var_9008_begin_0 = const()[name = tensor<string, []>("op_9008_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9008_end_0 = const()[name = tensor<string, []>("op_9008_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_9008_end_mask_0 = const()[name = tensor<string, []>("op_9008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9008_cast = slice_by_index(begin = var_9008_begin_0, end = var_9008_end_0, end_mask = var_9008_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9008_cast")];
            tensor<int32, [4]> var_9012_begin_0 = const()[name = tensor<string, []>("op_9012_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9012_end_0 = const()[name = tensor<string, []>("op_9012_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_9012_end_mask_0 = const()[name = tensor<string, []>("op_9012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9012_cast = slice_by_index(begin = var_9012_begin_0, end = var_9012_end_0, end_mask = var_9012_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9012_cast")];
            tensor<int32, [4]> var_9016_begin_0 = const()[name = tensor<string, []>("op_9016_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9016_end_0 = const()[name = tensor<string, []>("op_9016_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_9016_end_mask_0 = const()[name = tensor<string, []>("op_9016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9016_cast = slice_by_index(begin = var_9016_begin_0, end = var_9016_end_0, end_mask = var_9016_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9016_cast")];
            tensor<int32, [4]> var_9020_begin_0 = const()[name = tensor<string, []>("op_9020_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9020_end_0 = const()[name = tensor<string, []>("op_9020_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_9020_end_mask_0 = const()[name = tensor<string, []>("op_9020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9020_cast = slice_by_index(begin = var_9020_begin_0, end = var_9020_end_0, end_mask = var_9020_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9020_cast")];
            tensor<int32, [4]> var_9024_begin_0 = const()[name = tensor<string, []>("op_9024_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9024_end_0 = const()[name = tensor<string, []>("op_9024_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_9024_end_mask_0 = const()[name = tensor<string, []>("op_9024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9024_cast = slice_by_index(begin = var_9024_begin_0, end = var_9024_end_0, end_mask = var_9024_end_mask_0, x = v_35_cast)[name = tensor<string, []>("op_9024_cast")];
            tensor<string, []> var_9028_equation_0 = const()[name = tensor<string, []>("op_9028_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9028_cast = einsum(equation = var_9028_equation_0, values = (var_8870_cast, var_8787_cast))[name = tensor<string, []>("op_9028_cast")];
            tensor<fp16, []> var_9029_to_fp16 = const()[name = tensor<string, []>("op_9029_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_361_cast = mul(x = var_9028_cast, y = var_9029_to_fp16)[name = tensor<string, []>("aw_361_cast")];
            tensor<string, []> var_9032_equation_0 = const()[name = tensor<string, []>("op_9032_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9032_cast = einsum(equation = var_9032_equation_0, values = (var_8874_cast, var_8791_cast))[name = tensor<string, []>("op_9032_cast")];
            tensor<fp16, []> var_9033_to_fp16 = const()[name = tensor<string, []>("op_9033_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_363_cast = mul(x = var_9032_cast, y = var_9033_to_fp16)[name = tensor<string, []>("aw_363_cast")];
            tensor<string, []> var_9036_equation_0 = const()[name = tensor<string, []>("op_9036_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9036_cast = einsum(equation = var_9036_equation_0, values = (var_8878_cast, var_8795_cast))[name = tensor<string, []>("op_9036_cast")];
            tensor<fp16, []> var_9037_to_fp16 = const()[name = tensor<string, []>("op_9037_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_365_cast = mul(x = var_9036_cast, y = var_9037_to_fp16)[name = tensor<string, []>("aw_365_cast")];
            tensor<string, []> var_9040_equation_0 = const()[name = tensor<string, []>("op_9040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9040_cast = einsum(equation = var_9040_equation_0, values = (var_8882_cast, var_8799_cast))[name = tensor<string, []>("op_9040_cast")];
            tensor<fp16, []> var_9041_to_fp16 = const()[name = tensor<string, []>("op_9041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_367_cast = mul(x = var_9040_cast, y = var_9041_to_fp16)[name = tensor<string, []>("aw_367_cast")];
            tensor<string, []> var_9044_equation_0 = const()[name = tensor<string, []>("op_9044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9044_cast = einsum(equation = var_9044_equation_0, values = (var_8886_cast, var_8803_cast))[name = tensor<string, []>("op_9044_cast")];
            tensor<fp16, []> var_9045_to_fp16 = const()[name = tensor<string, []>("op_9045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_369_cast = mul(x = var_9044_cast, y = var_9045_to_fp16)[name = tensor<string, []>("aw_369_cast")];
            tensor<string, []> var_9048_equation_0 = const()[name = tensor<string, []>("op_9048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9048_cast = einsum(equation = var_9048_equation_0, values = (var_8890_cast, var_8807_cast))[name = tensor<string, []>("op_9048_cast")];
            tensor<fp16, []> var_9049_to_fp16 = const()[name = tensor<string, []>("op_9049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_371_cast = mul(x = var_9048_cast, y = var_9049_to_fp16)[name = tensor<string, []>("aw_371_cast")];
            tensor<string, []> var_9052_equation_0 = const()[name = tensor<string, []>("op_9052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9052_cast = einsum(equation = var_9052_equation_0, values = (var_8894_cast, var_8811_cast))[name = tensor<string, []>("op_9052_cast")];
            tensor<fp16, []> var_9053_to_fp16 = const()[name = tensor<string, []>("op_9053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_373_cast = mul(x = var_9052_cast, y = var_9053_to_fp16)[name = tensor<string, []>("aw_373_cast")];
            tensor<string, []> var_9056_equation_0 = const()[name = tensor<string, []>("op_9056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9056_cast = einsum(equation = var_9056_equation_0, values = (var_8898_cast, var_8815_cast))[name = tensor<string, []>("op_9056_cast")];
            tensor<fp16, []> var_9057_to_fp16 = const()[name = tensor<string, []>("op_9057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_375_cast = mul(x = var_9056_cast, y = var_9057_to_fp16)[name = tensor<string, []>("aw_375_cast")];
            tensor<string, []> var_9060_equation_0 = const()[name = tensor<string, []>("op_9060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9060_cast = einsum(equation = var_9060_equation_0, values = (var_8902_cast, var_8819_cast))[name = tensor<string, []>("op_9060_cast")];
            tensor<fp16, []> var_9061_to_fp16 = const()[name = tensor<string, []>("op_9061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_377_cast = mul(x = var_9060_cast, y = var_9061_to_fp16)[name = tensor<string, []>("aw_377_cast")];
            tensor<string, []> var_9064_equation_0 = const()[name = tensor<string, []>("op_9064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9064_cast = einsum(equation = var_9064_equation_0, values = (var_8906_cast, var_8823_cast))[name = tensor<string, []>("op_9064_cast")];
            tensor<fp16, []> var_9065_to_fp16 = const()[name = tensor<string, []>("op_9065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_379_cast = mul(x = var_9064_cast, y = var_9065_to_fp16)[name = tensor<string, []>("aw_379_cast")];
            tensor<string, []> var_9068_equation_0 = const()[name = tensor<string, []>("op_9068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9068_cast = einsum(equation = var_9068_equation_0, values = (var_8910_cast, var_8827_cast))[name = tensor<string, []>("op_9068_cast")];
            tensor<fp16, []> var_9069_to_fp16 = const()[name = tensor<string, []>("op_9069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_381_cast = mul(x = var_9068_cast, y = var_9069_to_fp16)[name = tensor<string, []>("aw_381_cast")];
            tensor<string, []> var_9072_equation_0 = const()[name = tensor<string, []>("op_9072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9072_cast = einsum(equation = var_9072_equation_0, values = (var_8914_cast, var_8831_cast))[name = tensor<string, []>("op_9072_cast")];
            tensor<fp16, []> var_9073_to_fp16 = const()[name = tensor<string, []>("op_9073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_383_cast = mul(x = var_9072_cast, y = var_9073_to_fp16)[name = tensor<string, []>("aw_383_cast")];
            tensor<string, []> var_9076_equation_0 = const()[name = tensor<string, []>("op_9076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9076_cast = einsum(equation = var_9076_equation_0, values = (var_8918_cast, var_8835_cast))[name = tensor<string, []>("op_9076_cast")];
            tensor<fp16, []> var_9077_to_fp16 = const()[name = tensor<string, []>("op_9077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_385_cast = mul(x = var_9076_cast, y = var_9077_to_fp16)[name = tensor<string, []>("aw_385_cast")];
            tensor<string, []> var_9080_equation_0 = const()[name = tensor<string, []>("op_9080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9080_cast = einsum(equation = var_9080_equation_0, values = (var_8922_cast, var_8839_cast))[name = tensor<string, []>("op_9080_cast")];
            tensor<fp16, []> var_9081_to_fp16 = const()[name = tensor<string, []>("op_9081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_387_cast = mul(x = var_9080_cast, y = var_9081_to_fp16)[name = tensor<string, []>("aw_387_cast")];
            tensor<string, []> var_9084_equation_0 = const()[name = tensor<string, []>("op_9084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9084_cast = einsum(equation = var_9084_equation_0, values = (var_8926_cast, var_8843_cast))[name = tensor<string, []>("op_9084_cast")];
            tensor<fp16, []> var_9085_to_fp16 = const()[name = tensor<string, []>("op_9085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_389_cast = mul(x = var_9084_cast, y = var_9085_to_fp16)[name = tensor<string, []>("aw_389_cast")];
            tensor<string, []> var_9088_equation_0 = const()[name = tensor<string, []>("op_9088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9088_cast = einsum(equation = var_9088_equation_0, values = (var_8930_cast, var_8847_cast))[name = tensor<string, []>("op_9088_cast")];
            tensor<fp16, []> var_9089_to_fp16 = const()[name = tensor<string, []>("op_9089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_391_cast = mul(x = var_9088_cast, y = var_9089_to_fp16)[name = tensor<string, []>("aw_391_cast")];
            tensor<string, []> var_9092_equation_0 = const()[name = tensor<string, []>("op_9092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9092_cast = einsum(equation = var_9092_equation_0, values = (var_8934_cast, var_8851_cast))[name = tensor<string, []>("op_9092_cast")];
            tensor<fp16, []> var_9093_to_fp16 = const()[name = tensor<string, []>("op_9093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_393_cast = mul(x = var_9092_cast, y = var_9093_to_fp16)[name = tensor<string, []>("aw_393_cast")];
            tensor<string, []> var_9096_equation_0 = const()[name = tensor<string, []>("op_9096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9096_cast = einsum(equation = var_9096_equation_0, values = (var_8938_cast, var_8855_cast))[name = tensor<string, []>("op_9096_cast")];
            tensor<fp16, []> var_9097_to_fp16 = const()[name = tensor<string, []>("op_9097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_395_cast = mul(x = var_9096_cast, y = var_9097_to_fp16)[name = tensor<string, []>("aw_395_cast")];
            tensor<string, []> var_9100_equation_0 = const()[name = tensor<string, []>("op_9100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9100_cast = einsum(equation = var_9100_equation_0, values = (var_8942_cast, var_8859_cast))[name = tensor<string, []>("op_9100_cast")];
            tensor<fp16, []> var_9101_to_fp16 = const()[name = tensor<string, []>("op_9101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_397_cast = mul(x = var_9100_cast, y = var_9101_to_fp16)[name = tensor<string, []>("aw_397_cast")];
            tensor<string, []> var_9104_equation_0 = const()[name = tensor<string, []>("op_9104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_9104_cast = einsum(equation = var_9104_equation_0, values = (var_8946_cast, var_8863_cast))[name = tensor<string, []>("op_9104_cast")];
            tensor<fp16, []> var_9105_to_fp16 = const()[name = tensor<string, []>("op_9105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_399_cast = mul(x = var_9104_cast, y = var_9105_to_fp16)[name = tensor<string, []>("aw_399_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9107_cast = softmax(axis = var_7228, x = aw_361_cast)[name = tensor<string, []>("op_9107_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9108_cast = softmax(axis = var_7228, x = aw_363_cast)[name = tensor<string, []>("op_9108_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9109_cast = softmax(axis = var_7228, x = aw_365_cast)[name = tensor<string, []>("op_9109_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9110_cast = softmax(axis = var_7228, x = aw_367_cast)[name = tensor<string, []>("op_9110_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9111_cast = softmax(axis = var_7228, x = aw_369_cast)[name = tensor<string, []>("op_9111_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9112_cast = softmax(axis = var_7228, x = aw_371_cast)[name = tensor<string, []>("op_9112_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9113_cast = softmax(axis = var_7228, x = aw_373_cast)[name = tensor<string, []>("op_9113_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9114_cast = softmax(axis = var_7228, x = aw_375_cast)[name = tensor<string, []>("op_9114_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9115_cast = softmax(axis = var_7228, x = aw_377_cast)[name = tensor<string, []>("op_9115_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9116_cast = softmax(axis = var_7228, x = aw_379_cast)[name = tensor<string, []>("op_9116_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9117_cast = softmax(axis = var_7228, x = aw_381_cast)[name = tensor<string, []>("op_9117_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9118_cast = softmax(axis = var_7228, x = aw_383_cast)[name = tensor<string, []>("op_9118_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9119_cast = softmax(axis = var_7228, x = aw_385_cast)[name = tensor<string, []>("op_9119_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9120_cast = softmax(axis = var_7228, x = aw_387_cast)[name = tensor<string, []>("op_9120_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9121_cast = softmax(axis = var_7228, x = aw_389_cast)[name = tensor<string, []>("op_9121_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9122_cast = softmax(axis = var_7228, x = aw_391_cast)[name = tensor<string, []>("op_9122_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9123_cast = softmax(axis = var_7228, x = aw_393_cast)[name = tensor<string, []>("op_9123_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9124_cast = softmax(axis = var_7228, x = aw_395_cast)[name = tensor<string, []>("op_9124_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9125_cast = softmax(axis = var_7228, x = aw_397_cast)[name = tensor<string, []>("op_9125_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_9126_cast = softmax(axis = var_7228, x = aw_399_cast)[name = tensor<string, []>("op_9126_cast")];
            tensor<string, []> var_9128_equation_0 = const()[name = tensor<string, []>("op_9128_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9128_cast = einsum(equation = var_9128_equation_0, values = (var_8948_cast, var_9107_cast))[name = tensor<string, []>("op_9128_cast")];
            tensor<string, []> var_9130_equation_0 = const()[name = tensor<string, []>("op_9130_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9130_cast = einsum(equation = var_9130_equation_0, values = (var_8952_cast, var_9108_cast))[name = tensor<string, []>("op_9130_cast")];
            tensor<string, []> var_9132_equation_0 = const()[name = tensor<string, []>("op_9132_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9132_cast = einsum(equation = var_9132_equation_0, values = (var_8956_cast, var_9109_cast))[name = tensor<string, []>("op_9132_cast")];
            tensor<string, []> var_9134_equation_0 = const()[name = tensor<string, []>("op_9134_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9134_cast = einsum(equation = var_9134_equation_0, values = (var_8960_cast, var_9110_cast))[name = tensor<string, []>("op_9134_cast")];
            tensor<string, []> var_9136_equation_0 = const()[name = tensor<string, []>("op_9136_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9136_cast = einsum(equation = var_9136_equation_0, values = (var_8964_cast, var_9111_cast))[name = tensor<string, []>("op_9136_cast")];
            tensor<string, []> var_9138_equation_0 = const()[name = tensor<string, []>("op_9138_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9138_cast = einsum(equation = var_9138_equation_0, values = (var_8968_cast, var_9112_cast))[name = tensor<string, []>("op_9138_cast")];
            tensor<string, []> var_9140_equation_0 = const()[name = tensor<string, []>("op_9140_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9140_cast = einsum(equation = var_9140_equation_0, values = (var_8972_cast, var_9113_cast))[name = tensor<string, []>("op_9140_cast")];
            tensor<string, []> var_9142_equation_0 = const()[name = tensor<string, []>("op_9142_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9142_cast = einsum(equation = var_9142_equation_0, values = (var_8976_cast, var_9114_cast))[name = tensor<string, []>("op_9142_cast")];
            tensor<string, []> var_9144_equation_0 = const()[name = tensor<string, []>("op_9144_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9144_cast = einsum(equation = var_9144_equation_0, values = (var_8980_cast, var_9115_cast))[name = tensor<string, []>("op_9144_cast")];
            tensor<string, []> var_9146_equation_0 = const()[name = tensor<string, []>("op_9146_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9146_cast = einsum(equation = var_9146_equation_0, values = (var_8984_cast, var_9116_cast))[name = tensor<string, []>("op_9146_cast")];
            tensor<string, []> var_9148_equation_0 = const()[name = tensor<string, []>("op_9148_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9148_cast = einsum(equation = var_9148_equation_0, values = (var_8988_cast, var_9117_cast))[name = tensor<string, []>("op_9148_cast")];
            tensor<string, []> var_9150_equation_0 = const()[name = tensor<string, []>("op_9150_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9150_cast = einsum(equation = var_9150_equation_0, values = (var_8992_cast, var_9118_cast))[name = tensor<string, []>("op_9150_cast")];
            tensor<string, []> var_9152_equation_0 = const()[name = tensor<string, []>("op_9152_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9152_cast = einsum(equation = var_9152_equation_0, values = (var_8996_cast, var_9119_cast))[name = tensor<string, []>("op_9152_cast")];
            tensor<string, []> var_9154_equation_0 = const()[name = tensor<string, []>("op_9154_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9154_cast = einsum(equation = var_9154_equation_0, values = (var_9000_cast, var_9120_cast))[name = tensor<string, []>("op_9154_cast")];
            tensor<string, []> var_9156_equation_0 = const()[name = tensor<string, []>("op_9156_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9156_cast = einsum(equation = var_9156_equation_0, values = (var_9004_cast, var_9121_cast))[name = tensor<string, []>("op_9156_cast")];
            tensor<string, []> var_9158_equation_0 = const()[name = tensor<string, []>("op_9158_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9158_cast = einsum(equation = var_9158_equation_0, values = (var_9008_cast, var_9122_cast))[name = tensor<string, []>("op_9158_cast")];
            tensor<string, []> var_9160_equation_0 = const()[name = tensor<string, []>("op_9160_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9160_cast = einsum(equation = var_9160_equation_0, values = (var_9012_cast, var_9123_cast))[name = tensor<string, []>("op_9160_cast")];
            tensor<string, []> var_9162_equation_0 = const()[name = tensor<string, []>("op_9162_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9162_cast = einsum(equation = var_9162_equation_0, values = (var_9016_cast, var_9124_cast))[name = tensor<string, []>("op_9162_cast")];
            tensor<string, []> var_9164_equation_0 = const()[name = tensor<string, []>("op_9164_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9164_cast = einsum(equation = var_9164_equation_0, values = (var_9020_cast, var_9125_cast))[name = tensor<string, []>("op_9164_cast")];
            tensor<string, []> var_9166_equation_0 = const()[name = tensor<string, []>("op_9166_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9166_cast = einsum(equation = var_9166_equation_0, values = (var_9024_cast, var_9126_cast))[name = tensor<string, []>("op_9166_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_7228, interleave = input_329_interleave_0, values = (var_9128_cast, var_9130_cast, var_9132_cast, var_9134_cast, var_9136_cast, var_9138_cast, var_9140_cast, var_9142_cast, var_9144_cast, var_9146_cast, var_9148_cast, var_9150_cast, var_9152_cast, var_9154_cast, var_9156_cast, var_9158_cast, var_9160_cast, var_9162_cast, var_9164_cast, var_9166_cast))[name = tensor<string, []>("input_329_cast")];
            tensor<int32, [2]> var_9172 = const()[name = tensor<string, []>("op_9172"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9174 = const()[name = tensor<string, []>("op_9174"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9176_pad_type_0 = const()[name = tensor<string, []>("op_9176_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9176_pad_0 = const()[name = tensor<string, []>("op_9176_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165438848))), lut = tensor<fp16, [4]>([-0x1.8acp-5, -0x1.d5cp-7, 0x1.d5cp-7, 0x1.8a8p-5]), 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, []>(165848512)))];
            tensor<fp16, [2, 1280, 1, 256]> var_9176_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_9174, groups = var_7228, pad = var_9176_pad_0, pad_type = var_9176_pad_type_0, strides = var_9172, 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_9176_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_53_cast = add(x = var_9176_cast, y = inputs_51_cast)[name = tensor<string, []>("inputs_53_cast")];
            tensor<int32, [1]> var_9180 = const()[name = tensor<string, []>("op_9180"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_53_cast = reduce_mean(axes = var_9180, keep_dims = var_7223, 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_9184 = const()[name = tensor<string, []>("op_9184"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_9185_cast = reduce_mean(axes = var_9184, keep_dims = var_7223, x = zero_mean_sq_53_cast)[name = tensor<string, []>("op_9185_cast")];
            tensor<fp16, []> var_9186_to_fp16 = const()[name = tensor<string, []>("op_9186_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_9187_cast = add(x = var_9185_cast, y = var_9186_to_fp16)[name = tensor<string, []>("op_9187_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_9187_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_9191_to_fp16 = const()[name = tensor<string, []>("op_9191_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165851136)))];
            tensor<fp16, [2, 1280, 1, 256]> var_9192_cast = add(x = out_53_cast, y = var_9191_to_fp16)[name = tensor<string, []>("op_9192_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9194_to_fp16 = const()[name = tensor<string, []>("op_9194_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165853760)))];
            tensor<fp16, [2, 1280, 1, 256]> input_331_cast = mul(x = var_9192_cast, y = var_9194_to_fp16)[name = tensor<string, []>("input_331_cast")];
            tensor<int32, [2]> var_9202 = const()[name = tensor<string, []>("op_9202"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9204 = const()[name = tensor<string, []>("op_9204"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9206_pad_type_0 = const()[name = tensor<string, []>("op_9206_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9206_pad_0 = const()[name = tensor<string, []>("op_9206_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, [3276800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165856384))), lut = tensor<fp16, [4]>([-0x1.6b4p-4, -0x1.ab8p-6, 0x1.a9p-6, 0x1.6a8p-4]), 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, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169133248))), lut = tensor<fp16, [4]>([-0x1.734p-3, -0x1.564p-4, -0x1.14cp-6, 0x1.d04p-6]), 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_9206_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_9204, groups = var_7228, pad = var_9206_pad_0, pad_type = var_9206_pad_type_0, strides = var_9202, 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_9206_cast")];
            tensor<int32, [2]> var_9207_split_sizes_0 = const()[name = tensor<string, []>("op_9207_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_9207_axis_0 = const()[name = tensor<string, []>("op_9207_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_9207_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_9207_cast_1 = split(axis = var_9207_axis_0, split_sizes = var_9207_split_sizes_0, x = var_9206_cast)[name = tensor<string, []>("op_9207_cast")];
            tensor<string, []> var_9209_mode_0 = const()[name = tensor<string, []>("op_9209_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_9209_cast = gelu(mode = var_9209_mode_0, x = var_9207_cast_1)[name = tensor<string, []>("op_9209_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_333_cast = mul(x = var_9207_cast_0, y = var_9209_cast)[name = tensor<string, []>("input_333_cast")];
            tensor<int32, [2]> var_9213 = const()[name = tensor<string, []>("op_9213"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9215 = const()[name = tensor<string, []>("op_9215"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9217_pad_type_0 = const()[name = tensor<string, []>("op_9217_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9217_pad_0 = const()[name = tensor<string, []>("op_9217_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, [1638400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169135872))), lut = tensor<fp16, [4]>([-0x1.6bp-4, -0x1.b0cp-6, 0x1.aep-6, 0x1.6ap-4]), 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, []>(170774336)))];
            tensor<fp16, [2, 1280, 1, 256]> var_9217_cast = conv(bias = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_9215, groups = var_7228, pad = var_9217_pad_0, pad_type = var_9217_pad_type_0, strides = var_9213, weight = up_blocks_1_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_333_cast)[name = tensor<string, []>("op_9217_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_193_cast = add(x = var_9217_cast, y = inputs_53_cast)[name = tensor<string, []>("hidden_states_193_cast")];
            tensor<int32, [4]> var_9219 = const()[name = tensor<string, []>("op_9219"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_335_cast = reshape(shape = var_9219, x = hidden_states_193_cast)[name = tensor<string, []>("input_335_cast")];
            tensor<int32, [2]> var_9223 = const()[name = tensor<string, []>("op_9223"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9225 = const()[name = tensor<string, []>("op_9225"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(170776960))), lut = tensor<fp16, [4]>([-0x1.60cp-4, -0x1.a2cp-6, 0x1.abp-6, 0x1.624p-4]), 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, []>(171186624)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_195_cast = conv(bias = up_blocks_1_attentions_1_proj_out_bias_to_fp16, dilations = var_9225, groups = var_7228, pad = hidden_states_195_pad_0, pad_type = hidden_states_195_pad_type_0, strides = var_9223, 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_7228, 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, []>(171189248)))];
            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, []>(171193152)))];
            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, []>(171197056)))];
            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, []>(171200960)))];
            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_9243 = const()[name = tensor<string, []>("op_9243"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9245 = const()[name = tensor<string, []>("op_9245"), 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, [5529600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171204864))), lut = tensor<fp16, [4]>([-0x1.254p-4, -0x1.46p-6, 0x1.3e8p-6, 0x1.23p-4]), 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, []>(176734528)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_199_cast = conv(bias = up_blocks_1_resnets_2_conv1_bias_to_fp16, dilations = var_9245, groups = var_7228, pad = hidden_states_199_pad_0, pad_type = hidden_states_199_pad_type_0, strides = var_9243, 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_9251 = const()[name = tensor<string, []>("op_9251"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9253 = const()[name = tensor<string, []>("op_9253"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176737152))), lut = tensor<fp16, [4]>([-0x1.22p-4, -0x1.acp-9, 0x1.ap-9, 0x1.b3p-5]), 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, []>(177146816)))];
            tensor<fp16, [2, 1280, 1, 1]> temb_31_cast = conv(bias = up_blocks_1_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_9253, groups = var_7228, pad = temb_31_pad_0, pad_type = temb_31_pad_type_0, strides = var_9251, 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, []>(177149440)))];
            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, []>(177152064)))];
            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_9263 = const()[name = tensor<string, []>("op_9263"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9265 = const()[name = tensor<string, []>("op_9265"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(177154688))), lut = tensor<fp16, [4]>([-0x1.578p-4, -0x1.84cp-6, 0x1.864p-6, 0x1.584p-4]), 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, []>(180841152)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_201_cast = conv(bias = up_blocks_1_resnets_2_conv2_bias_to_fp16, dilations = var_9265, groups = var_7228, pad = hidden_states_201_pad_0, pad_type = hidden_states_201_pad_type_0, strides = var_9263, 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_9270 = const()[name = tensor<string, []>("op_9270"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9272 = const()[name = tensor<string, []>("op_9272"), 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, [614400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180843776))), lut = tensor<fp16, [4]>([-0x1.92cp-6, -0x1.bdcp-8, 0x1.bc8p-8, 0x1.928p-6]), 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, []>(181458240)))];
            tensor<fp16, [2, 1280, 16, 16]> x_15_cast = conv(bias = up_blocks_1_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_9272, groups = var_7228, pad = x_15_pad_0, pad_type = x_15_pad_type_0, strides = var_9270, 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, []>(181460864)))];
            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, []>(181463488)))];
            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_9292 = const()[name = tensor<string, []>("op_9292"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9294 = const()[name = tensor<string, []>("op_9294"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181466112))), lut = tensor<fp16, [4]>([-0x1.99p-4, -0x1.edp-6, 0x1.f0cp-6, 0x1.9ap-4]), 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, []>(181875776)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_205_cast = conv(bias = up_blocks_1_attentions_2_proj_in_bias_to_fp16, dilations = var_9294, groups = var_7228, pad = hidden_states_205_pad_0, pad_type = hidden_states_205_pad_type_0, strides = var_9292, 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_9299 = const()[name = tensor<string, []>("op_9299"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<fp16, [2, 1280, 1, 256]> inputs_55_cast = reshape(shape = var_9299, x = hidden_states_205_cast)[name = tensor<string, []>("inputs_55_cast")];
            tensor<int32, [1]> var_9309 = const()[name = tensor<string, []>("op_9309"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_55_cast = reduce_mean(axes = var_9309, keep_dims = var_7223, 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_9313 = const()[name = tensor<string, []>("op_9313"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_9314_cast = reduce_mean(axes = var_9313, keep_dims = var_7223, x = zero_mean_sq_55_cast)[name = tensor<string, []>("op_9314_cast")];
            tensor<fp16, []> var_9315_to_fp16 = const()[name = tensor<string, []>("op_9315_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_9316_cast = add(x = var_9314_cast, y = var_9315_to_fp16)[name = tensor<string, []>("op_9316_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_9316_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_9320_to_fp16 = const()[name = tensor<string, []>("op_9320_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181878400)))];
            tensor<fp16, [2, 1280, 1, 256]> var_9321_cast = add(x = out_55_cast, y = var_9320_to_fp16)[name = tensor<string, []>("op_9321_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9323_to_fp16 = const()[name = tensor<string, []>("op_9323_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181881024)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_207_cast = mul(x = var_9321_cast, y = var_9323_to_fp16)[name = tensor<string, []>("hidden_states_207_cast")];
            tensor<int32, [2]> var_9330 = const()[name = tensor<string, []>("op_9330"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9332 = const()[name = tensor<string, []>("op_9332"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181883648))), lut = tensor<fp16, [4]>([-0x1.9cp-4, -0x1.eb8p-6, 0x1.ec4p-6, 0x1.9c4p-4]), 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_9332, groups = var_7228, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = var_9330, 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_9336 = const()[name = tensor<string, []>("op_9336"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9338 = const()[name = tensor<string, []>("op_9338"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(182293312))), lut = tensor<fp16, [4]>([-0x1.a04p-4, -0x1.f24p-6, 0x1.ec8p-6, 0x1.9e8p-4]), 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_9338, groups = var_7228, pad = k_73_pad_0, pad_type = k_73_pad_type_0, strides = var_9336, 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_9342 = const()[name = tensor<string, []>("op_9342"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9344 = const()[name = tensor<string, []>("op_9344"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(182702976))), lut = tensor<fp16, [4]>([-0x1.408p-4, -0x1.7ep-6, 0x1.7f4p-6, 0x1.408p-4]), 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_9344, groups = var_7228, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = var_9342, 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_9348_begin_0 = const()[name = tensor<string, []>("op_9348_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9348_end_0 = const()[name = tensor<string, []>("op_9348_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_9348_end_mask_0 = const()[name = tensor<string, []>("op_9348_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9348_cast = slice_by_index(begin = var_9348_begin_0, end = var_9348_end_0, end_mask = var_9348_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9348_cast")];
            tensor<int32, [4]> var_9352_begin_0 = const()[name = tensor<string, []>("op_9352_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9352_end_0 = const()[name = tensor<string, []>("op_9352_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_9352_end_mask_0 = const()[name = tensor<string, []>("op_9352_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9352_cast = slice_by_index(begin = var_9352_begin_0, end = var_9352_end_0, end_mask = var_9352_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9352_cast")];
            tensor<int32, [4]> var_9356_begin_0 = const()[name = tensor<string, []>("op_9356_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9356_end_0 = const()[name = tensor<string, []>("op_9356_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_9356_end_mask_0 = const()[name = tensor<string, []>("op_9356_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9356_cast = slice_by_index(begin = var_9356_begin_0, end = var_9356_end_0, end_mask = var_9356_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9356_cast")];
            tensor<int32, [4]> var_9360_begin_0 = const()[name = tensor<string, []>("op_9360_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9360_end_0 = const()[name = tensor<string, []>("op_9360_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_9360_end_mask_0 = const()[name = tensor<string, []>("op_9360_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9360_cast = slice_by_index(begin = var_9360_begin_0, end = var_9360_end_0, end_mask = var_9360_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9360_cast")];
            tensor<int32, [4]> var_9364_begin_0 = const()[name = tensor<string, []>("op_9364_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9364_end_0 = const()[name = tensor<string, []>("op_9364_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_9364_end_mask_0 = const()[name = tensor<string, []>("op_9364_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9364_cast = slice_by_index(begin = var_9364_begin_0, end = var_9364_end_0, end_mask = var_9364_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9364_cast")];
            tensor<int32, [4]> var_9368_begin_0 = const()[name = tensor<string, []>("op_9368_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9368_end_0 = const()[name = tensor<string, []>("op_9368_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_9368_end_mask_0 = const()[name = tensor<string, []>("op_9368_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9368_cast = slice_by_index(begin = var_9368_begin_0, end = var_9368_end_0, end_mask = var_9368_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9368_cast")];
            tensor<int32, [4]> var_9372_begin_0 = const()[name = tensor<string, []>("op_9372_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9372_end_0 = const()[name = tensor<string, []>("op_9372_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_9372_end_mask_0 = const()[name = tensor<string, []>("op_9372_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9372_cast = slice_by_index(begin = var_9372_begin_0, end = var_9372_end_0, end_mask = var_9372_end_mask_0, x = q_37_cast)[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, 448, 0, 0])];
            tensor<int32, [4]> var_9376_end_0 = const()[name = tensor<string, []>("op_9376_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_9376_end_mask_0 = const()[name = tensor<string, []>("op_9376_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9376_cast = slice_by_index(begin = var_9376_begin_0, end = var_9376_end_0, end_mask = var_9376_end_mask_0, x = q_37_cast)[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, 512, 0, 0])];
            tensor<int32, [4]> var_9380_end_0 = const()[name = tensor<string, []>("op_9380_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_9380_end_mask_0 = const()[name = tensor<string, []>("op_9380_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9380_cast = slice_by_index(begin = var_9380_begin_0, end = var_9380_end_0, end_mask = var_9380_end_mask_0, x = q_37_cast)[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, 576, 0, 0])];
            tensor<int32, [4]> var_9384_end_0 = const()[name = tensor<string, []>("op_9384_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_9384_end_mask_0 = const()[name = tensor<string, []>("op_9384_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9384_cast = slice_by_index(begin = var_9384_begin_0, end = var_9384_end_0, end_mask = var_9384_end_mask_0, x = q_37_cast)[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, 640, 0, 0])];
            tensor<int32, [4]> var_9388_end_0 = const()[name = tensor<string, []>("op_9388_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_9388_end_mask_0 = const()[name = tensor<string, []>("op_9388_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9388_cast = slice_by_index(begin = var_9388_begin_0, end = var_9388_end_0, end_mask = var_9388_end_mask_0, x = q_37_cast)[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, 704, 0, 0])];
            tensor<int32, [4]> var_9392_end_0 = const()[name = tensor<string, []>("op_9392_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_9392_end_mask_0 = const()[name = tensor<string, []>("op_9392_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9392_cast = slice_by_index(begin = var_9392_begin_0, end = var_9392_end_0, end_mask = var_9392_end_mask_0, x = q_37_cast)[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, 768, 0, 0])];
            tensor<int32, [4]> var_9396_end_0 = const()[name = tensor<string, []>("op_9396_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_9396_end_mask_0 = const()[name = tensor<string, []>("op_9396_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9396_cast = slice_by_index(begin = var_9396_begin_0, end = var_9396_end_0, end_mask = var_9396_end_mask_0, x = q_37_cast)[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, 832, 0, 0])];
            tensor<int32, [4]> var_9400_end_0 = const()[name = tensor<string, []>("op_9400_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_9400_end_mask_0 = const()[name = tensor<string, []>("op_9400_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9400_cast = slice_by_index(begin = var_9400_begin_0, end = var_9400_end_0, end_mask = var_9400_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9400_cast")];
            tensor<int32, [4]> var_9404_begin_0 = const()[name = tensor<string, []>("op_9404_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9404_end_0 = const()[name = tensor<string, []>("op_9404_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_9404_end_mask_0 = const()[name = tensor<string, []>("op_9404_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9404_cast = slice_by_index(begin = var_9404_begin_0, end = var_9404_end_0, end_mask = var_9404_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9404_cast")];
            tensor<int32, [4]> var_9408_begin_0 = const()[name = tensor<string, []>("op_9408_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9408_end_0 = const()[name = tensor<string, []>("op_9408_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_9408_end_mask_0 = const()[name = tensor<string, []>("op_9408_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9408_cast = slice_by_index(begin = var_9408_begin_0, end = var_9408_end_0, end_mask = var_9408_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9408_cast")];
            tensor<int32, [4]> var_9412_begin_0 = const()[name = tensor<string, []>("op_9412_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9412_end_0 = const()[name = tensor<string, []>("op_9412_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_9412_end_mask_0 = const()[name = tensor<string, []>("op_9412_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9412_cast = slice_by_index(begin = var_9412_begin_0, end = var_9412_end_0, end_mask = var_9412_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9412_cast")];
            tensor<int32, [4]> var_9416_begin_0 = const()[name = tensor<string, []>("op_9416_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9416_end_0 = const()[name = tensor<string, []>("op_9416_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_9416_end_mask_0 = const()[name = tensor<string, []>("op_9416_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9416_cast = slice_by_index(begin = var_9416_begin_0, end = var_9416_end_0, end_mask = var_9416_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9416_cast")];
            tensor<int32, [4]> var_9420_begin_0 = const()[name = tensor<string, []>("op_9420_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9420_end_0 = const()[name = tensor<string, []>("op_9420_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_9420_end_mask_0 = const()[name = tensor<string, []>("op_9420_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9420_cast = slice_by_index(begin = var_9420_begin_0, end = var_9420_end_0, end_mask = var_9420_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9420_cast")];
            tensor<int32, [4]> var_9424_begin_0 = const()[name = tensor<string, []>("op_9424_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9424_end_0 = const()[name = tensor<string, []>("op_9424_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_9424_end_mask_0 = const()[name = tensor<string, []>("op_9424_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9424_cast = slice_by_index(begin = var_9424_begin_0, end = var_9424_end_0, end_mask = var_9424_end_mask_0, x = q_37_cast)[name = tensor<string, []>("op_9424_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_9431_begin_0 = const()[name = tensor<string, []>("op_9431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9431_end_0 = const()[name = tensor<string, []>("op_9431_end_0"), val = tensor<int32, [4]>([2, 256, 1, 64])];
            tensor<bool, [4]> var_9431_end_mask_0 = const()[name = tensor<string, []>("op_9431_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, 64]> var_9431_cast = slice_by_index(begin = var_9431_begin_0, end = var_9431_end_0, end_mask = var_9431_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9431_cast")];
            tensor<int32, [4]> var_9435_begin_0 = const()[name = tensor<string, []>("op_9435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9435_end_0 = const()[name = tensor<string, []>("op_9435_end_0"), val = tensor<int32, [4]>([2, 256, 1, 128])];
            tensor<bool, [4]> var_9435_end_mask_0 = const()[name = tensor<string, []>("op_9435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9435_cast = slice_by_index(begin = var_9435_begin_0, end = var_9435_end_0, end_mask = var_9435_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9435_cast")];
            tensor<int32, [4]> var_9439_begin_0 = const()[name = tensor<string, []>("op_9439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9439_end_0 = const()[name = tensor<string, []>("op_9439_end_0"), val = tensor<int32, [4]>([2, 256, 1, 192])];
            tensor<bool, [4]> var_9439_end_mask_0 = const()[name = tensor<string, []>("op_9439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9439_cast = slice_by_index(begin = var_9439_begin_0, end = var_9439_end_0, end_mask = var_9439_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9439_cast")];
            tensor<int32, [4]> var_9443_begin_0 = const()[name = tensor<string, []>("op_9443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9443_end_0 = const()[name = tensor<string, []>("op_9443_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_9443_end_mask_0 = const()[name = tensor<string, []>("op_9443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9443_cast = slice_by_index(begin = var_9443_begin_0, end = var_9443_end_0, end_mask = var_9443_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9443_cast")];
            tensor<int32, [4]> var_9447_begin_0 = const()[name = tensor<string, []>("op_9447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9447_end_0 = const()[name = tensor<string, []>("op_9447_end_0"), val = tensor<int32, [4]>([2, 256, 1, 320])];
            tensor<bool, [4]> var_9447_end_mask_0 = const()[name = tensor<string, []>("op_9447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9447_cast = slice_by_index(begin = var_9447_begin_0, end = var_9447_end_0, end_mask = var_9447_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9447_cast")];
            tensor<int32, [4]> var_9451_begin_0 = const()[name = tensor<string, []>("op_9451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9451_end_0 = const()[name = tensor<string, []>("op_9451_end_0"), val = tensor<int32, [4]>([2, 256, 1, 384])];
            tensor<bool, [4]> var_9451_end_mask_0 = const()[name = tensor<string, []>("op_9451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9451_cast = slice_by_index(begin = var_9451_begin_0, end = var_9451_end_0, end_mask = var_9451_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9451_cast")];
            tensor<int32, [4]> var_9455_begin_0 = const()[name = tensor<string, []>("op_9455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9455_end_0 = const()[name = tensor<string, []>("op_9455_end_0"), val = tensor<int32, [4]>([2, 256, 1, 448])];
            tensor<bool, [4]> var_9455_end_mask_0 = const()[name = tensor<string, []>("op_9455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9455_cast = slice_by_index(begin = var_9455_begin_0, end = var_9455_end_0, end_mask = var_9455_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9455_cast")];
            tensor<int32, [4]> var_9459_begin_0 = const()[name = tensor<string, []>("op_9459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9459_end_0 = const()[name = tensor<string, []>("op_9459_end_0"), val = tensor<int32, [4]>([2, 256, 1, 512])];
            tensor<bool, [4]> var_9459_end_mask_0 = const()[name = tensor<string, []>("op_9459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9459_cast = slice_by_index(begin = var_9459_begin_0, end = var_9459_end_0, end_mask = var_9459_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9459_cast")];
            tensor<int32, [4]> var_9463_begin_0 = const()[name = tensor<string, []>("op_9463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9463_end_0 = const()[name = tensor<string, []>("op_9463_end_0"), val = tensor<int32, [4]>([2, 256, 1, 576])];
            tensor<bool, [4]> var_9463_end_mask_0 = const()[name = tensor<string, []>("op_9463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9463_cast = slice_by_index(begin = var_9463_begin_0, end = var_9463_end_0, end_mask = var_9463_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9463_cast")];
            tensor<int32, [4]> var_9467_begin_0 = const()[name = tensor<string, []>("op_9467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9467_end_0 = const()[name = tensor<string, []>("op_9467_end_0"), val = tensor<int32, [4]>([2, 256, 1, 640])];
            tensor<bool, [4]> var_9467_end_mask_0 = const()[name = tensor<string, []>("op_9467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9467_cast = slice_by_index(begin = var_9467_begin_0, end = var_9467_end_0, end_mask = var_9467_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9467_cast")];
            tensor<int32, [4]> var_9471_begin_0 = const()[name = tensor<string, []>("op_9471_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_9471_end_0 = const()[name = tensor<string, []>("op_9471_end_0"), val = tensor<int32, [4]>([2, 256, 1, 704])];
            tensor<bool, [4]> var_9471_end_mask_0 = const()[name = tensor<string, []>("op_9471_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9471_cast = slice_by_index(begin = var_9471_begin_0, end = var_9471_end_0, end_mask = var_9471_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9471_cast")];
            tensor<int32, [4]> var_9475_begin_0 = const()[name = tensor<string, []>("op_9475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_9475_end_0 = const()[name = tensor<string, []>("op_9475_end_0"), val = tensor<int32, [4]>([2, 256, 1, 768])];
            tensor<bool, [4]> var_9475_end_mask_0 = const()[name = tensor<string, []>("op_9475_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9475_cast = slice_by_index(begin = var_9475_begin_0, end = var_9475_end_0, end_mask = var_9475_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9475_cast")];
            tensor<int32, [4]> var_9479_begin_0 = const()[name = tensor<string, []>("op_9479_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_9479_end_0 = const()[name = tensor<string, []>("op_9479_end_0"), val = tensor<int32, [4]>([2, 256, 1, 832])];
            tensor<bool, [4]> var_9479_end_mask_0 = const()[name = tensor<string, []>("op_9479_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9479_cast = slice_by_index(begin = var_9479_begin_0, end = var_9479_end_0, end_mask = var_9479_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9479_cast")];
            tensor<int32, [4]> var_9483_begin_0 = const()[name = tensor<string, []>("op_9483_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_9483_end_0 = const()[name = tensor<string, []>("op_9483_end_0"), val = tensor<int32, [4]>([2, 256, 1, 896])];
            tensor<bool, [4]> var_9483_end_mask_0 = const()[name = tensor<string, []>("op_9483_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9483_cast = slice_by_index(begin = var_9483_begin_0, end = var_9483_end_0, end_mask = var_9483_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9483_cast")];
            tensor<int32, [4]> var_9487_begin_0 = const()[name = tensor<string, []>("op_9487_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_9487_end_0 = const()[name = tensor<string, []>("op_9487_end_0"), val = tensor<int32, [4]>([2, 256, 1, 960])];
            tensor<bool, [4]> var_9487_end_mask_0 = const()[name = tensor<string, []>("op_9487_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9487_cast = slice_by_index(begin = var_9487_begin_0, end = var_9487_end_0, end_mask = var_9487_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9487_cast")];
            tensor<int32, [4]> var_9491_begin_0 = const()[name = tensor<string, []>("op_9491_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_9491_end_0 = const()[name = tensor<string, []>("op_9491_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_9491_end_mask_0 = const()[name = tensor<string, []>("op_9491_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9491_cast = slice_by_index(begin = var_9491_begin_0, end = var_9491_end_0, end_mask = var_9491_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9491_cast")];
            tensor<int32, [4]> var_9495_begin_0 = const()[name = tensor<string, []>("op_9495_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_9495_end_0 = const()[name = tensor<string, []>("op_9495_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1088])];
            tensor<bool, [4]> var_9495_end_mask_0 = const()[name = tensor<string, []>("op_9495_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9495_cast = slice_by_index(begin = var_9495_begin_0, end = var_9495_end_0, end_mask = var_9495_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9495_cast")];
            tensor<int32, [4]> var_9499_begin_0 = const()[name = tensor<string, []>("op_9499_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_9499_end_0 = const()[name = tensor<string, []>("op_9499_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1152])];
            tensor<bool, [4]> var_9499_end_mask_0 = const()[name = tensor<string, []>("op_9499_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9499_cast = slice_by_index(begin = var_9499_begin_0, end = var_9499_end_0, end_mask = var_9499_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9499_cast")];
            tensor<int32, [4]> var_9503_begin_0 = const()[name = tensor<string, []>("op_9503_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_9503_end_0 = const()[name = tensor<string, []>("op_9503_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1216])];
            tensor<bool, [4]> var_9503_end_mask_0 = const()[name = tensor<string, []>("op_9503_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9503_cast = slice_by_index(begin = var_9503_begin_0, end = var_9503_end_0, end_mask = var_9503_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9503_cast")];
            tensor<int32, [4]> var_9507_begin_0 = const()[name = tensor<string, []>("op_9507_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_9507_end_0 = const()[name = tensor<string, []>("op_9507_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1280])];
            tensor<bool, [4]> var_9507_end_mask_0 = const()[name = tensor<string, []>("op_9507_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 256, 1, 64]> var_9507_cast = slice_by_index(begin = var_9507_begin_0, end = var_9507_end_0, end_mask = var_9507_end_mask_0, x = transpose_13)[name = tensor<string, []>("op_9507_cast")];
            tensor<int32, [4]> var_9509_begin_0 = const()[name = tensor<string, []>("op_9509_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9509_end_0 = const()[name = tensor<string, []>("op_9509_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_9509_end_mask_0 = const()[name = tensor<string, []>("op_9509_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9509_cast = slice_by_index(begin = var_9509_begin_0, end = var_9509_end_0, end_mask = var_9509_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9509_cast")];
            tensor<int32, [4]> var_9513_begin_0 = const()[name = tensor<string, []>("op_9513_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9513_end_0 = const()[name = tensor<string, []>("op_9513_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_9513_end_mask_0 = const()[name = tensor<string, []>("op_9513_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9513_cast = slice_by_index(begin = var_9513_begin_0, end = var_9513_end_0, end_mask = var_9513_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9513_cast")];
            tensor<int32, [4]> var_9517_begin_0 = const()[name = tensor<string, []>("op_9517_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9517_end_0 = const()[name = tensor<string, []>("op_9517_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_9517_end_mask_0 = const()[name = tensor<string, []>("op_9517_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9517_cast = slice_by_index(begin = var_9517_begin_0, end = var_9517_end_0, end_mask = var_9517_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9517_cast")];
            tensor<int32, [4]> var_9521_begin_0 = const()[name = tensor<string, []>("op_9521_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9521_end_0 = const()[name = tensor<string, []>("op_9521_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_9521_end_mask_0 = const()[name = tensor<string, []>("op_9521_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9521_cast = slice_by_index(begin = var_9521_begin_0, end = var_9521_end_0, end_mask = var_9521_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9521_cast")];
            tensor<int32, [4]> var_9525_begin_0 = const()[name = tensor<string, []>("op_9525_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9525_end_0 = const()[name = tensor<string, []>("op_9525_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_9525_end_mask_0 = const()[name = tensor<string, []>("op_9525_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9525_cast = slice_by_index(begin = var_9525_begin_0, end = var_9525_end_0, end_mask = var_9525_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9525_cast")];
            tensor<int32, [4]> var_9529_begin_0 = const()[name = tensor<string, []>("op_9529_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9529_end_0 = const()[name = tensor<string, []>("op_9529_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_9529_end_mask_0 = const()[name = tensor<string, []>("op_9529_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9529_cast = slice_by_index(begin = var_9529_begin_0, end = var_9529_end_0, end_mask = var_9529_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9529_cast")];
            tensor<int32, [4]> var_9533_begin_0 = const()[name = tensor<string, []>("op_9533_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9533_end_0 = const()[name = tensor<string, []>("op_9533_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_9533_end_mask_0 = const()[name = tensor<string, []>("op_9533_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9533_cast = slice_by_index(begin = var_9533_begin_0, end = var_9533_end_0, end_mask = var_9533_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9533_cast")];
            tensor<int32, [4]> var_9537_begin_0 = const()[name = tensor<string, []>("op_9537_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9537_end_0 = const()[name = tensor<string, []>("op_9537_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_9537_end_mask_0 = const()[name = tensor<string, []>("op_9537_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9537_cast = slice_by_index(begin = var_9537_begin_0, end = var_9537_end_0, end_mask = var_9537_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9537_cast")];
            tensor<int32, [4]> var_9541_begin_0 = const()[name = tensor<string, []>("op_9541_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9541_end_0 = const()[name = tensor<string, []>("op_9541_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_9541_end_mask_0 = const()[name = tensor<string, []>("op_9541_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9541_cast = slice_by_index(begin = var_9541_begin_0, end = var_9541_end_0, end_mask = var_9541_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9541_cast")];
            tensor<int32, [4]> var_9545_begin_0 = const()[name = tensor<string, []>("op_9545_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9545_end_0 = const()[name = tensor<string, []>("op_9545_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_9545_end_mask_0 = const()[name = tensor<string, []>("op_9545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9545_cast = slice_by_index(begin = var_9545_begin_0, end = var_9545_end_0, end_mask = var_9545_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9545_cast")];
            tensor<int32, [4]> var_9549_begin_0 = const()[name = tensor<string, []>("op_9549_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9549_end_0 = const()[name = tensor<string, []>("op_9549_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_9549_end_mask_0 = const()[name = tensor<string, []>("op_9549_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9549_cast = slice_by_index(begin = var_9549_begin_0, end = var_9549_end_0, end_mask = var_9549_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9549_cast")];
            tensor<int32, [4]> var_9553_begin_0 = const()[name = tensor<string, []>("op_9553_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9553_end_0 = const()[name = tensor<string, []>("op_9553_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_9553_end_mask_0 = const()[name = tensor<string, []>("op_9553_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9553_cast = slice_by_index(begin = var_9553_begin_0, end = var_9553_end_0, end_mask = var_9553_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9553_cast")];
            tensor<int32, [4]> var_9557_begin_0 = const()[name = tensor<string, []>("op_9557_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9557_end_0 = const()[name = tensor<string, []>("op_9557_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_9557_end_mask_0 = const()[name = tensor<string, []>("op_9557_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9557_cast = slice_by_index(begin = var_9557_begin_0, end = var_9557_end_0, end_mask = var_9557_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9557_cast")];
            tensor<int32, [4]> var_9561_begin_0 = const()[name = tensor<string, []>("op_9561_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9561_end_0 = const()[name = tensor<string, []>("op_9561_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_9561_end_mask_0 = const()[name = tensor<string, []>("op_9561_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9561_cast = slice_by_index(begin = var_9561_begin_0, end = var_9561_end_0, end_mask = var_9561_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9561_cast")];
            tensor<int32, [4]> var_9565_begin_0 = const()[name = tensor<string, []>("op_9565_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9565_end_0 = const()[name = tensor<string, []>("op_9565_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_9565_end_mask_0 = const()[name = tensor<string, []>("op_9565_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9565_cast = slice_by_index(begin = var_9565_begin_0, end = var_9565_end_0, end_mask = var_9565_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9565_cast")];
            tensor<int32, [4]> var_9569_begin_0 = const()[name = tensor<string, []>("op_9569_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9569_end_0 = const()[name = tensor<string, []>("op_9569_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_9569_end_mask_0 = const()[name = tensor<string, []>("op_9569_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9569_cast = slice_by_index(begin = var_9569_begin_0, end = var_9569_end_0, end_mask = var_9569_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9569_cast")];
            tensor<int32, [4]> var_9573_begin_0 = const()[name = tensor<string, []>("op_9573_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9573_end_0 = const()[name = tensor<string, []>("op_9573_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_9573_end_mask_0 = const()[name = tensor<string, []>("op_9573_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9573_cast = slice_by_index(begin = var_9573_begin_0, end = var_9573_end_0, end_mask = var_9573_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9573_cast")];
            tensor<int32, [4]> var_9577_begin_0 = const()[name = tensor<string, []>("op_9577_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9577_end_0 = const()[name = tensor<string, []>("op_9577_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_9577_end_mask_0 = const()[name = tensor<string, []>("op_9577_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9577_cast = slice_by_index(begin = var_9577_begin_0, end = var_9577_end_0, end_mask = var_9577_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9577_cast")];
            tensor<int32, [4]> var_9581_begin_0 = const()[name = tensor<string, []>("op_9581_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9581_end_0 = const()[name = tensor<string, []>("op_9581_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_9581_end_mask_0 = const()[name = tensor<string, []>("op_9581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9581_cast = slice_by_index(begin = var_9581_begin_0, end = var_9581_end_0, end_mask = var_9581_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9581_cast")];
            tensor<int32, [4]> var_9585_begin_0 = const()[name = tensor<string, []>("op_9585_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9585_end_0 = const()[name = tensor<string, []>("op_9585_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_9585_end_mask_0 = const()[name = tensor<string, []>("op_9585_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9585_cast = slice_by_index(begin = var_9585_begin_0, end = var_9585_end_0, end_mask = var_9585_end_mask_0, x = v_37_cast)[name = tensor<string, []>("op_9585_cast")];
            tensor<string, []> var_9589_equation_0 = const()[name = tensor<string, []>("op_9589_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9589_cast = einsum(equation = var_9589_equation_0, values = (var_9431_cast, var_9348_cast))[name = tensor<string, []>("op_9589_cast")];
            tensor<fp16, []> var_9590_to_fp16 = const()[name = tensor<string, []>("op_9590_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_401_cast = mul(x = var_9589_cast, y = var_9590_to_fp16)[name = tensor<string, []>("aw_401_cast")];
            tensor<string, []> var_9593_equation_0 = const()[name = tensor<string, []>("op_9593_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9593_cast = einsum(equation = var_9593_equation_0, values = (var_9435_cast, var_9352_cast))[name = tensor<string, []>("op_9593_cast")];
            tensor<fp16, []> var_9594_to_fp16 = const()[name = tensor<string, []>("op_9594_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_403_cast = mul(x = var_9593_cast, y = var_9594_to_fp16)[name = tensor<string, []>("aw_403_cast")];
            tensor<string, []> var_9597_equation_0 = const()[name = tensor<string, []>("op_9597_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9597_cast = einsum(equation = var_9597_equation_0, values = (var_9439_cast, var_9356_cast))[name = tensor<string, []>("op_9597_cast")];
            tensor<fp16, []> var_9598_to_fp16 = const()[name = tensor<string, []>("op_9598_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_405_cast = mul(x = var_9597_cast, y = var_9598_to_fp16)[name = tensor<string, []>("aw_405_cast")];
            tensor<string, []> var_9601_equation_0 = const()[name = tensor<string, []>("op_9601_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9601_cast = einsum(equation = var_9601_equation_0, values = (var_9443_cast, var_9360_cast))[name = tensor<string, []>("op_9601_cast")];
            tensor<fp16, []> var_9602_to_fp16 = const()[name = tensor<string, []>("op_9602_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_407_cast = mul(x = var_9601_cast, y = var_9602_to_fp16)[name = tensor<string, []>("aw_407_cast")];
            tensor<string, []> var_9605_equation_0 = const()[name = tensor<string, []>("op_9605_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9605_cast = einsum(equation = var_9605_equation_0, values = (var_9447_cast, var_9364_cast))[name = tensor<string, []>("op_9605_cast")];
            tensor<fp16, []> var_9606_to_fp16 = const()[name = tensor<string, []>("op_9606_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_409_cast = mul(x = var_9605_cast, y = var_9606_to_fp16)[name = tensor<string, []>("aw_409_cast")];
            tensor<string, []> var_9609_equation_0 = const()[name = tensor<string, []>("op_9609_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9609_cast = einsum(equation = var_9609_equation_0, values = (var_9451_cast, var_9368_cast))[name = tensor<string, []>("op_9609_cast")];
            tensor<fp16, []> var_9610_to_fp16 = const()[name = tensor<string, []>("op_9610_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_411_cast = mul(x = var_9609_cast, y = var_9610_to_fp16)[name = tensor<string, []>("aw_411_cast")];
            tensor<string, []> var_9613_equation_0 = const()[name = tensor<string, []>("op_9613_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9613_cast = einsum(equation = var_9613_equation_0, values = (var_9455_cast, var_9372_cast))[name = tensor<string, []>("op_9613_cast")];
            tensor<fp16, []> var_9614_to_fp16 = const()[name = tensor<string, []>("op_9614_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_413_cast = mul(x = var_9613_cast, y = var_9614_to_fp16)[name = tensor<string, []>("aw_413_cast")];
            tensor<string, []> var_9617_equation_0 = const()[name = tensor<string, []>("op_9617_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9617_cast = einsum(equation = var_9617_equation_0, values = (var_9459_cast, var_9376_cast))[name = tensor<string, []>("op_9617_cast")];
            tensor<fp16, []> var_9618_to_fp16 = const()[name = tensor<string, []>("op_9618_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_415_cast = mul(x = var_9617_cast, y = var_9618_to_fp16)[name = tensor<string, []>("aw_415_cast")];
            tensor<string, []> var_9621_equation_0 = const()[name = tensor<string, []>("op_9621_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9621_cast = einsum(equation = var_9621_equation_0, values = (var_9463_cast, var_9380_cast))[name = tensor<string, []>("op_9621_cast")];
            tensor<fp16, []> var_9622_to_fp16 = const()[name = tensor<string, []>("op_9622_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_417_cast = mul(x = var_9621_cast, y = var_9622_to_fp16)[name = tensor<string, []>("aw_417_cast")];
            tensor<string, []> var_9625_equation_0 = const()[name = tensor<string, []>("op_9625_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9625_cast = einsum(equation = var_9625_equation_0, values = (var_9467_cast, var_9384_cast))[name = tensor<string, []>("op_9625_cast")];
            tensor<fp16, []> var_9626_to_fp16 = const()[name = tensor<string, []>("op_9626_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_419_cast = mul(x = var_9625_cast, y = var_9626_to_fp16)[name = tensor<string, []>("aw_419_cast")];
            tensor<string, []> var_9629_equation_0 = const()[name = tensor<string, []>("op_9629_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9629_cast = einsum(equation = var_9629_equation_0, values = (var_9471_cast, var_9388_cast))[name = tensor<string, []>("op_9629_cast")];
            tensor<fp16, []> var_9630_to_fp16 = const()[name = tensor<string, []>("op_9630_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_421_cast = mul(x = var_9629_cast, y = var_9630_to_fp16)[name = tensor<string, []>("aw_421_cast")];
            tensor<string, []> var_9633_equation_0 = const()[name = tensor<string, []>("op_9633_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9633_cast = einsum(equation = var_9633_equation_0, values = (var_9475_cast, var_9392_cast))[name = tensor<string, []>("op_9633_cast")];
            tensor<fp16, []> var_9634_to_fp16 = const()[name = tensor<string, []>("op_9634_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_423_cast = mul(x = var_9633_cast, y = var_9634_to_fp16)[name = tensor<string, []>("aw_423_cast")];
            tensor<string, []> var_9637_equation_0 = const()[name = tensor<string, []>("op_9637_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9637_cast = einsum(equation = var_9637_equation_0, values = (var_9479_cast, var_9396_cast))[name = tensor<string, []>("op_9637_cast")];
            tensor<fp16, []> var_9638_to_fp16 = const()[name = tensor<string, []>("op_9638_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_425_cast = mul(x = var_9637_cast, y = var_9638_to_fp16)[name = tensor<string, []>("aw_425_cast")];
            tensor<string, []> var_9641_equation_0 = const()[name = tensor<string, []>("op_9641_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9641_cast = einsum(equation = var_9641_equation_0, values = (var_9483_cast, var_9400_cast))[name = tensor<string, []>("op_9641_cast")];
            tensor<fp16, []> var_9642_to_fp16 = const()[name = tensor<string, []>("op_9642_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_427_cast = mul(x = var_9641_cast, y = var_9642_to_fp16)[name = tensor<string, []>("aw_427_cast")];
            tensor<string, []> var_9645_equation_0 = const()[name = tensor<string, []>("op_9645_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9645_cast = einsum(equation = var_9645_equation_0, values = (var_9487_cast, var_9404_cast))[name = tensor<string, []>("op_9645_cast")];
            tensor<fp16, []> var_9646_to_fp16 = const()[name = tensor<string, []>("op_9646_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_429_cast = mul(x = var_9645_cast, y = var_9646_to_fp16)[name = tensor<string, []>("aw_429_cast")];
            tensor<string, []> var_9649_equation_0 = const()[name = tensor<string, []>("op_9649_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9649_cast = einsum(equation = var_9649_equation_0, values = (var_9491_cast, var_9408_cast))[name = tensor<string, []>("op_9649_cast")];
            tensor<fp16, []> var_9650_to_fp16 = const()[name = tensor<string, []>("op_9650_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_431_cast = mul(x = var_9649_cast, y = var_9650_to_fp16)[name = tensor<string, []>("aw_431_cast")];
            tensor<string, []> var_9653_equation_0 = const()[name = tensor<string, []>("op_9653_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9653_cast = einsum(equation = var_9653_equation_0, values = (var_9495_cast, var_9412_cast))[name = tensor<string, []>("op_9653_cast")];
            tensor<fp16, []> var_9654_to_fp16 = const()[name = tensor<string, []>("op_9654_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_433_cast = mul(x = var_9653_cast, y = var_9654_to_fp16)[name = tensor<string, []>("aw_433_cast")];
            tensor<string, []> var_9657_equation_0 = const()[name = tensor<string, []>("op_9657_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9657_cast = einsum(equation = var_9657_equation_0, values = (var_9499_cast, var_9416_cast))[name = tensor<string, []>("op_9657_cast")];
            tensor<fp16, []> var_9658_to_fp16 = const()[name = tensor<string, []>("op_9658_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_435_cast = mul(x = var_9657_cast, y = var_9658_to_fp16)[name = tensor<string, []>("aw_435_cast")];
            tensor<string, []> var_9661_equation_0 = const()[name = tensor<string, []>("op_9661_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9661_cast = einsum(equation = var_9661_equation_0, values = (var_9503_cast, var_9420_cast))[name = tensor<string, []>("op_9661_cast")];
            tensor<fp16, []> var_9662_to_fp16 = const()[name = tensor<string, []>("op_9662_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_437_cast = mul(x = var_9661_cast, y = var_9662_to_fp16)[name = tensor<string, []>("aw_437_cast")];
            tensor<string, []> var_9665_equation_0 = const()[name = tensor<string, []>("op_9665_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 256, 1, 256]> var_9665_cast = einsum(equation = var_9665_equation_0, values = (var_9507_cast, var_9424_cast))[name = tensor<string, []>("op_9665_cast")];
            tensor<fp16, []> var_9666_to_fp16 = const()[name = tensor<string, []>("op_9666_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 256, 1, 256]> aw_439_cast = mul(x = var_9665_cast, y = var_9666_to_fp16)[name = tensor<string, []>("aw_439_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9668_cast = softmax(axis = var_7228, x = aw_401_cast)[name = tensor<string, []>("op_9668_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9669_cast = softmax(axis = var_7228, x = aw_403_cast)[name = tensor<string, []>("op_9669_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9670_cast = softmax(axis = var_7228, x = aw_405_cast)[name = tensor<string, []>("op_9670_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9671_cast = softmax(axis = var_7228, x = aw_407_cast)[name = tensor<string, []>("op_9671_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9672_cast = softmax(axis = var_7228, x = aw_409_cast)[name = tensor<string, []>("op_9672_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9673_cast = softmax(axis = var_7228, x = aw_411_cast)[name = tensor<string, []>("op_9673_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9674_cast = softmax(axis = var_7228, x = aw_413_cast)[name = tensor<string, []>("op_9674_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9675_cast = softmax(axis = var_7228, x = aw_415_cast)[name = tensor<string, []>("op_9675_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9676_cast = softmax(axis = var_7228, x = aw_417_cast)[name = tensor<string, []>("op_9676_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9677_cast = softmax(axis = var_7228, x = aw_419_cast)[name = tensor<string, []>("op_9677_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9678_cast = softmax(axis = var_7228, x = aw_421_cast)[name = tensor<string, []>("op_9678_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9679_cast = softmax(axis = var_7228, x = aw_423_cast)[name = tensor<string, []>("op_9679_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9680_cast = softmax(axis = var_7228, x = aw_425_cast)[name = tensor<string, []>("op_9680_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9681_cast = softmax(axis = var_7228, x = aw_427_cast)[name = tensor<string, []>("op_9681_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9682_cast = softmax(axis = var_7228, x = aw_429_cast)[name = tensor<string, []>("op_9682_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9683_cast = softmax(axis = var_7228, x = aw_431_cast)[name = tensor<string, []>("op_9683_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9684_cast = softmax(axis = var_7228, x = aw_433_cast)[name = tensor<string, []>("op_9684_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9685_cast = softmax(axis = var_7228, x = aw_435_cast)[name = tensor<string, []>("op_9685_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9686_cast = softmax(axis = var_7228, x = aw_437_cast)[name = tensor<string, []>("op_9686_cast")];
            tensor<fp16, [2, 256, 1, 256]> var_9687_cast = softmax(axis = var_7228, x = aw_439_cast)[name = tensor<string, []>("op_9687_cast")];
            tensor<string, []> var_9689_equation_0 = const()[name = tensor<string, []>("op_9689_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9689_cast = einsum(equation = var_9689_equation_0, values = (var_9509_cast, var_9668_cast))[name = tensor<string, []>("op_9689_cast")];
            tensor<string, []> var_9691_equation_0 = const()[name = tensor<string, []>("op_9691_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9691_cast = einsum(equation = var_9691_equation_0, values = (var_9513_cast, var_9669_cast))[name = tensor<string, []>("op_9691_cast")];
            tensor<string, []> var_9693_equation_0 = const()[name = tensor<string, []>("op_9693_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9693_cast = einsum(equation = var_9693_equation_0, values = (var_9517_cast, var_9670_cast))[name = tensor<string, []>("op_9693_cast")];
            tensor<string, []> var_9695_equation_0 = const()[name = tensor<string, []>("op_9695_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9695_cast = einsum(equation = var_9695_equation_0, values = (var_9521_cast, var_9671_cast))[name = tensor<string, []>("op_9695_cast")];
            tensor<string, []> var_9697_equation_0 = const()[name = tensor<string, []>("op_9697_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9697_cast = einsum(equation = var_9697_equation_0, values = (var_9525_cast, var_9672_cast))[name = tensor<string, []>("op_9697_cast")];
            tensor<string, []> var_9699_equation_0 = const()[name = tensor<string, []>("op_9699_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9699_cast = einsum(equation = var_9699_equation_0, values = (var_9529_cast, var_9673_cast))[name = tensor<string, []>("op_9699_cast")];
            tensor<string, []> var_9701_equation_0 = const()[name = tensor<string, []>("op_9701_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9701_cast = einsum(equation = var_9701_equation_0, values = (var_9533_cast, var_9674_cast))[name = tensor<string, []>("op_9701_cast")];
            tensor<string, []> var_9703_equation_0 = const()[name = tensor<string, []>("op_9703_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9703_cast = einsum(equation = var_9703_equation_0, values = (var_9537_cast, var_9675_cast))[name = tensor<string, []>("op_9703_cast")];
            tensor<string, []> var_9705_equation_0 = const()[name = tensor<string, []>("op_9705_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9705_cast = einsum(equation = var_9705_equation_0, values = (var_9541_cast, var_9676_cast))[name = tensor<string, []>("op_9705_cast")];
            tensor<string, []> var_9707_equation_0 = const()[name = tensor<string, []>("op_9707_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9707_cast = einsum(equation = var_9707_equation_0, values = (var_9545_cast, var_9677_cast))[name = tensor<string, []>("op_9707_cast")];
            tensor<string, []> var_9709_equation_0 = const()[name = tensor<string, []>("op_9709_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9709_cast = einsum(equation = var_9709_equation_0, values = (var_9549_cast, var_9678_cast))[name = tensor<string, []>("op_9709_cast")];
            tensor<string, []> var_9711_equation_0 = const()[name = tensor<string, []>("op_9711_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9711_cast = einsum(equation = var_9711_equation_0, values = (var_9553_cast, var_9679_cast))[name = tensor<string, []>("op_9711_cast")];
            tensor<string, []> var_9713_equation_0 = const()[name = tensor<string, []>("op_9713_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9713_cast = einsum(equation = var_9713_equation_0, values = (var_9557_cast, var_9680_cast))[name = tensor<string, []>("op_9713_cast")];
            tensor<string, []> var_9715_equation_0 = const()[name = tensor<string, []>("op_9715_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9715_cast = einsum(equation = var_9715_equation_0, values = (var_9561_cast, var_9681_cast))[name = tensor<string, []>("op_9715_cast")];
            tensor<string, []> var_9717_equation_0 = const()[name = tensor<string, []>("op_9717_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9717_cast = einsum(equation = var_9717_equation_0, values = (var_9565_cast, var_9682_cast))[name = tensor<string, []>("op_9717_cast")];
            tensor<string, []> var_9719_equation_0 = const()[name = tensor<string, []>("op_9719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9719_cast = einsum(equation = var_9719_equation_0, values = (var_9569_cast, var_9683_cast))[name = tensor<string, []>("op_9719_cast")];
            tensor<string, []> var_9721_equation_0 = const()[name = tensor<string, []>("op_9721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9721_cast = einsum(equation = var_9721_equation_0, values = (var_9573_cast, var_9684_cast))[name = tensor<string, []>("op_9721_cast")];
            tensor<string, []> var_9723_equation_0 = const()[name = tensor<string, []>("op_9723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9723_cast = einsum(equation = var_9723_equation_0, values = (var_9577_cast, var_9685_cast))[name = tensor<string, []>("op_9723_cast")];
            tensor<string, []> var_9725_equation_0 = const()[name = tensor<string, []>("op_9725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9725_cast = einsum(equation = var_9725_equation_0, values = (var_9581_cast, var_9686_cast))[name = tensor<string, []>("op_9725_cast")];
            tensor<string, []> var_9727_equation_0 = const()[name = tensor<string, []>("op_9727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_9727_cast = einsum(equation = var_9727_equation_0, values = (var_9585_cast, var_9687_cast))[name = tensor<string, []>("op_9727_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_7228, interleave = input_353_interleave_0, values = (var_9689_cast, var_9691_cast, var_9693_cast, var_9695_cast, var_9697_cast, var_9699_cast, var_9701_cast, var_9703_cast, var_9705_cast, var_9707_cast, var_9709_cast, var_9711_cast, var_9713_cast, var_9715_cast, var_9717_cast, var_9719_cast, var_9721_cast, var_9723_cast, var_9725_cast, var_9727_cast))[name = tensor<string, []>("input_353_cast")];
            tensor<int32, [2]> var_9733 = const()[name = tensor<string, []>("op_9733"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9735 = const()[name = tensor<string, []>("op_9735"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_9737_pad_type_0 = const()[name = tensor<string, []>("op_9737_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_9737_pad_0 = const()[name = tensor<string, []>("op_9737_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183112640))), lut = tensor<fp16, [4]>([-0x1.558p-4, -0x1.99cp-6, 0x1.95p-6, 0x1.544p-4]), 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, []>(183522304)))];
            tensor<fp16, [2, 1280, 1, 256]> var_9737_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_9735, groups = var_7228, pad = var_9737_pad_0, pad_type = var_9737_pad_type_0, strides = var_9733, 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_9737_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_57_cast = add(x = var_9737_cast, y = inputs_55_cast)[name = tensor<string, []>("inputs_57_cast")];
            tensor<int32, [1]> var_9741 = const()[name = tensor<string, []>("op_9741"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_57_cast = reduce_mean(axes = var_9741, keep_dims = var_7223, 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_9745 = const()[name = tensor<string, []>("op_9745"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_9746_cast = reduce_mean(axes = var_9745, keep_dims = var_7223, x = zero_mean_sq_57_cast)[name = tensor<string, []>("op_9746_cast")];
            tensor<fp16, []> var_9747_to_fp16 = const()[name = tensor<string, []>("op_9747_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_9748_cast = add(x = var_9746_cast, y = var_9747_to_fp16)[name = tensor<string, []>("op_9748_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_9748_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_9752_to_fp16 = const()[name = tensor<string, []>("op_9752_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183524928)))];
            tensor<fp16, [2, 1280, 1, 256]> var_9753_cast = add(x = out_57_cast, y = var_9752_to_fp16)[name = tensor<string, []>("op_9753_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_9755_to_fp16 = const()[name = tensor<string, []>("op_9755_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183527552)))];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_209_cast = mul(x = var_9753_cast, y = var_9755_to_fp16)[name = tensor<string, []>("hidden_states_209_cast")];
            tensor<int32, [2]> var_9762 = const()[name = tensor<string, []>("op_9762"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9764 = const()[name = tensor<string, []>("op_9764"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183530176))), lut = tensor<fp16, [4]>([-0x1.45p-4, -0x1.854p-6, 0x1.85p-6, 0x1.454p-4]), 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_9764, groups = var_7228, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = var_9762, 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_9768 = const()[name = tensor<string, []>("op_9768"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9770 = const()[name = tensor<string, []>("op_9770"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_77_pad_type_0 = const()[name = tensor<string, []>("k_77_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_77_pad_0 = const()[name = tensor<string, []>("k_77_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183939840))), lut = tensor<fp16, [4]>([-0x1.544p-4, -0x1.96p-6, 0x1.8f4p-6, 0x1.528p-4]), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> k_77_cast = conv(dilations = var_9770, groups = var_7228, pad = k_77_pad_0, pad_type = k_77_pad_type_0, strides = var_9768, 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_9774 = const()[name = tensor<string, []>("op_9774"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_9776 = const()[name = tensor<string, []>("op_9776"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_39_pad_type_0 = const()[name = tensor<string, []>("v_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_39_pad_0 = const()[name = tensor<string, []>("v_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [1280, 1024, 1, 1]> up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [327680]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184267584))), lut = tensor<fp16, [4]>([-0x1.09cp-4, -0x1.378p-6, 0x1.364p-6, 0x1.094p-4]), name = tensor<string, []>("up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([1280, 1024, 1, 1])];
            tensor<fp16, [2, 1280, 1, 77]> v_39_cast = conv(dilations = var_9776, groups = var_7228, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = var_9774, 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_9780_begin_0 = const()[name = tensor<string, []>("op_9780_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9780_end_0 = const()[name = tensor<string, []>("op_9780_end_0"), val = tensor<int32, [4]>([2, 64, 1, 256])];
            tensor<bool, [4]> var_9780_end_mask_0 = const()[name = tensor<string, []>("op_9780_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9780_cast = slice_by_index(begin = var_9780_begin_0, end = var_9780_end_0, end_mask = var_9780_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9780_cast")];
            tensor<int32, [4]> var_9784_begin_0 = const()[name = tensor<string, []>("op_9784_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9784_end_0 = const()[name = tensor<string, []>("op_9784_end_0"), val = tensor<int32, [4]>([2, 128, 1, 256])];
            tensor<bool, [4]> var_9784_end_mask_0 = const()[name = tensor<string, []>("op_9784_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9784_cast = slice_by_index(begin = var_9784_begin_0, end = var_9784_end_0, end_mask = var_9784_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9784_cast")];
            tensor<int32, [4]> var_9788_begin_0 = const()[name = tensor<string, []>("op_9788_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9788_end_0 = const()[name = tensor<string, []>("op_9788_end_0"), val = tensor<int32, [4]>([2, 192, 1, 256])];
            tensor<bool, [4]> var_9788_end_mask_0 = const()[name = tensor<string, []>("op_9788_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9788_cast = slice_by_index(begin = var_9788_begin_0, end = var_9788_end_0, end_mask = var_9788_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9788_cast")];
            tensor<int32, [4]> var_9792_begin_0 = const()[name = tensor<string, []>("op_9792_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9792_end_0 = const()[name = tensor<string, []>("op_9792_end_0"), val = tensor<int32, [4]>([2, 256, 1, 256])];
            tensor<bool, [4]> var_9792_end_mask_0 = const()[name = tensor<string, []>("op_9792_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9792_cast = slice_by_index(begin = var_9792_begin_0, end = var_9792_end_0, end_mask = var_9792_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9792_cast")];
            tensor<int32, [4]> var_9796_begin_0 = const()[name = tensor<string, []>("op_9796_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9796_end_0 = const()[name = tensor<string, []>("op_9796_end_0"), val = tensor<int32, [4]>([2, 320, 1, 256])];
            tensor<bool, [4]> var_9796_end_mask_0 = const()[name = tensor<string, []>("op_9796_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9796_cast = slice_by_index(begin = var_9796_begin_0, end = var_9796_end_0, end_mask = var_9796_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9796_cast")];
            tensor<int32, [4]> var_9800_begin_0 = const()[name = tensor<string, []>("op_9800_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9800_end_0 = const()[name = tensor<string, []>("op_9800_end_0"), val = tensor<int32, [4]>([2, 384, 1, 256])];
            tensor<bool, [4]> var_9800_end_mask_0 = const()[name = tensor<string, []>("op_9800_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9800_cast = slice_by_index(begin = var_9800_begin_0, end = var_9800_end_0, end_mask = var_9800_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9800_cast")];
            tensor<int32, [4]> var_9804_begin_0 = const()[name = tensor<string, []>("op_9804_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9804_end_0 = const()[name = tensor<string, []>("op_9804_end_0"), val = tensor<int32, [4]>([2, 448, 1, 256])];
            tensor<bool, [4]> var_9804_end_mask_0 = const()[name = tensor<string, []>("op_9804_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9804_cast = slice_by_index(begin = var_9804_begin_0, end = var_9804_end_0, end_mask = var_9804_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9804_cast")];
            tensor<int32, [4]> var_9808_begin_0 = const()[name = tensor<string, []>("op_9808_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9808_end_0 = const()[name = tensor<string, []>("op_9808_end_0"), val = tensor<int32, [4]>([2, 512, 1, 256])];
            tensor<bool, [4]> var_9808_end_mask_0 = const()[name = tensor<string, []>("op_9808_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9808_cast = slice_by_index(begin = var_9808_begin_0, end = var_9808_end_0, end_mask = var_9808_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9808_cast")];
            tensor<int32, [4]> var_9812_begin_0 = const()[name = tensor<string, []>("op_9812_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9812_end_0 = const()[name = tensor<string, []>("op_9812_end_0"), val = tensor<int32, [4]>([2, 576, 1, 256])];
            tensor<bool, [4]> var_9812_end_mask_0 = const()[name = tensor<string, []>("op_9812_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9812_cast = slice_by_index(begin = var_9812_begin_0, end = var_9812_end_0, end_mask = var_9812_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9812_cast")];
            tensor<int32, [4]> var_9816_begin_0 = const()[name = tensor<string, []>("op_9816_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9816_end_0 = const()[name = tensor<string, []>("op_9816_end_0"), val = tensor<int32, [4]>([2, 640, 1, 256])];
            tensor<bool, [4]> var_9816_end_mask_0 = const()[name = tensor<string, []>("op_9816_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9816_cast = slice_by_index(begin = var_9816_begin_0, end = var_9816_end_0, end_mask = var_9816_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9816_cast")];
            tensor<int32, [4]> var_9820_begin_0 = const()[name = tensor<string, []>("op_9820_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9820_end_0 = const()[name = tensor<string, []>("op_9820_end_0"), val = tensor<int32, [4]>([2, 704, 1, 256])];
            tensor<bool, [4]> var_9820_end_mask_0 = const()[name = tensor<string, []>("op_9820_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9820_cast = slice_by_index(begin = var_9820_begin_0, end = var_9820_end_0, end_mask = var_9820_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9820_cast")];
            tensor<int32, [4]> var_9824_begin_0 = const()[name = tensor<string, []>("op_9824_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9824_end_0 = const()[name = tensor<string, []>("op_9824_end_0"), val = tensor<int32, [4]>([2, 768, 1, 256])];
            tensor<bool, [4]> var_9824_end_mask_0 = const()[name = tensor<string, []>("op_9824_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9824_cast = slice_by_index(begin = var_9824_begin_0, end = var_9824_end_0, end_mask = var_9824_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9824_cast")];
            tensor<int32, [4]> var_9828_begin_0 = const()[name = tensor<string, []>("op_9828_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9828_end_0 = const()[name = tensor<string, []>("op_9828_end_0"), val = tensor<int32, [4]>([2, 832, 1, 256])];
            tensor<bool, [4]> var_9828_end_mask_0 = const()[name = tensor<string, []>("op_9828_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9828_cast = slice_by_index(begin = var_9828_begin_0, end = var_9828_end_0, end_mask = var_9828_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9828_cast")];
            tensor<int32, [4]> var_9832_begin_0 = const()[name = tensor<string, []>("op_9832_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9832_end_0 = const()[name = tensor<string, []>("op_9832_end_0"), val = tensor<int32, [4]>([2, 896, 1, 256])];
            tensor<bool, [4]> var_9832_end_mask_0 = const()[name = tensor<string, []>("op_9832_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9832_cast = slice_by_index(begin = var_9832_begin_0, end = var_9832_end_0, end_mask = var_9832_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9832_cast")];
            tensor<int32, [4]> var_9836_begin_0 = const()[name = tensor<string, []>("op_9836_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9836_end_0 = const()[name = tensor<string, []>("op_9836_end_0"), val = tensor<int32, [4]>([2, 960, 1, 256])];
            tensor<bool, [4]> var_9836_end_mask_0 = const()[name = tensor<string, []>("op_9836_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9836_cast = slice_by_index(begin = var_9836_begin_0, end = var_9836_end_0, end_mask = var_9836_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9836_cast")];
            tensor<int32, [4]> var_9840_begin_0 = const()[name = tensor<string, []>("op_9840_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_9840_end_0 = const()[name = tensor<string, []>("op_9840_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_9840_end_mask_0 = const()[name = tensor<string, []>("op_9840_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9840_cast = slice_by_index(begin = var_9840_begin_0, end = var_9840_end_0, end_mask = var_9840_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9840_cast")];
            tensor<int32, [4]> var_9844_begin_0 = const()[name = tensor<string, []>("op_9844_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_9844_end_0 = const()[name = tensor<string, []>("op_9844_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 256])];
            tensor<bool, [4]> var_9844_end_mask_0 = const()[name = tensor<string, []>("op_9844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9844_cast = slice_by_index(begin = var_9844_begin_0, end = var_9844_end_0, end_mask = var_9844_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9844_cast")];
            tensor<int32, [4]> var_9848_begin_0 = const()[name = tensor<string, []>("op_9848_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_9848_end_0 = const()[name = tensor<string, []>("op_9848_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 256])];
            tensor<bool, [4]> var_9848_end_mask_0 = const()[name = tensor<string, []>("op_9848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9848_cast = slice_by_index(begin = var_9848_begin_0, end = var_9848_end_0, end_mask = var_9848_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9848_cast")];
            tensor<int32, [4]> var_9852_begin_0 = const()[name = tensor<string, []>("op_9852_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_9852_end_0 = const()[name = tensor<string, []>("op_9852_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 256])];
            tensor<bool, [4]> var_9852_end_mask_0 = const()[name = tensor<string, []>("op_9852_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9852_cast = slice_by_index(begin = var_9852_begin_0, end = var_9852_end_0, end_mask = var_9852_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9852_cast")];
            tensor<int32, [4]> var_9856_begin_0 = const()[name = tensor<string, []>("op_9856_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_9856_end_0 = const()[name = tensor<string, []>("op_9856_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 256])];
            tensor<bool, [4]> var_9856_end_mask_0 = const()[name = tensor<string, []>("op_9856_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 256]> var_9856_cast = slice_by_index(begin = var_9856_begin_0, end = var_9856_end_0, end_mask = var_9856_end_mask_0, x = q_39_cast)[name = tensor<string, []>("op_9856_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_9863_begin_0 = const()[name = tensor<string, []>("op_9863_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9863_end_0 = const()[name = tensor<string, []>("op_9863_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_9863_end_mask_0 = const()[name = tensor<string, []>("op_9863_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, 64]> var_9863_cast = slice_by_index(begin = var_9863_begin_0, end = var_9863_end_0, end_mask = var_9863_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9863_cast")];
            tensor<int32, [4]> var_9867_begin_0 = const()[name = tensor<string, []>("op_9867_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_9867_end_0 = const()[name = tensor<string, []>("op_9867_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_9867_end_mask_0 = const()[name = tensor<string, []>("op_9867_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9867_cast = slice_by_index(begin = var_9867_begin_0, end = var_9867_end_0, end_mask = var_9867_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9867_cast")];
            tensor<int32, [4]> var_9871_begin_0 = const()[name = tensor<string, []>("op_9871_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_9871_end_0 = const()[name = tensor<string, []>("op_9871_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_9871_end_mask_0 = const()[name = tensor<string, []>("op_9871_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9871_cast = slice_by_index(begin = var_9871_begin_0, end = var_9871_end_0, end_mask = var_9871_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9871_cast")];
            tensor<int32, [4]> var_9875_begin_0 = const()[name = tensor<string, []>("op_9875_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_9875_end_0 = const()[name = tensor<string, []>("op_9875_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_9875_end_mask_0 = const()[name = tensor<string, []>("op_9875_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9875_cast = slice_by_index(begin = var_9875_begin_0, end = var_9875_end_0, end_mask = var_9875_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9875_cast")];
            tensor<int32, [4]> var_9879_begin_0 = const()[name = tensor<string, []>("op_9879_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_9879_end_0 = const()[name = tensor<string, []>("op_9879_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_9879_end_mask_0 = const()[name = tensor<string, []>("op_9879_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9879_cast = slice_by_index(begin = var_9879_begin_0, end = var_9879_end_0, end_mask = var_9879_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9879_cast")];
            tensor<int32, [4]> var_9883_begin_0 = const()[name = tensor<string, []>("op_9883_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_9883_end_0 = const()[name = tensor<string, []>("op_9883_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_9883_end_mask_0 = const()[name = tensor<string, []>("op_9883_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9883_cast = slice_by_index(begin = var_9883_begin_0, end = var_9883_end_0, end_mask = var_9883_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9883_cast")];
            tensor<int32, [4]> var_9887_begin_0 = const()[name = tensor<string, []>("op_9887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_9887_end_0 = const()[name = tensor<string, []>("op_9887_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_9887_end_mask_0 = const()[name = tensor<string, []>("op_9887_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9887_cast = slice_by_index(begin = var_9887_begin_0, end = var_9887_end_0, end_mask = var_9887_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9887_cast")];
            tensor<int32, [4]> var_9891_begin_0 = const()[name = tensor<string, []>("op_9891_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_9891_end_0 = const()[name = tensor<string, []>("op_9891_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_9891_end_mask_0 = const()[name = tensor<string, []>("op_9891_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9891_cast = slice_by_index(begin = var_9891_begin_0, end = var_9891_end_0, end_mask = var_9891_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9891_cast")];
            tensor<int32, [4]> var_9895_begin_0 = const()[name = tensor<string, []>("op_9895_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_9895_end_0 = const()[name = tensor<string, []>("op_9895_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_9895_end_mask_0 = const()[name = tensor<string, []>("op_9895_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9895_cast = slice_by_index(begin = var_9895_begin_0, end = var_9895_end_0, end_mask = var_9895_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9895_cast")];
            tensor<int32, [4]> var_9899_begin_0 = const()[name = tensor<string, []>("op_9899_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_9899_end_0 = const()[name = tensor<string, []>("op_9899_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_9899_end_mask_0 = const()[name = tensor<string, []>("op_9899_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9899_cast = slice_by_index(begin = var_9899_begin_0, end = var_9899_end_0, end_mask = var_9899_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9899_cast")];
            tensor<int32, [4]> var_9903_begin_0 = const()[name = tensor<string, []>("op_9903_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 640])];
            tensor<int32, [4]> var_9903_end_0 = const()[name = tensor<string, []>("op_9903_end_0"), val = tensor<int32, [4]>([2, 77, 1, 704])];
            tensor<bool, [4]> var_9903_end_mask_0 = const()[name = tensor<string, []>("op_9903_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9903_cast = slice_by_index(begin = var_9903_begin_0, end = var_9903_end_0, end_mask = var_9903_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9903_cast")];
            tensor<int32, [4]> var_9907_begin_0 = const()[name = tensor<string, []>("op_9907_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 704])];
            tensor<int32, [4]> var_9907_end_0 = const()[name = tensor<string, []>("op_9907_end_0"), val = tensor<int32, [4]>([2, 77, 1, 768])];
            tensor<bool, [4]> var_9907_end_mask_0 = const()[name = tensor<string, []>("op_9907_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9907_cast = slice_by_index(begin = var_9907_begin_0, end = var_9907_end_0, end_mask = var_9907_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9907_cast")];
            tensor<int32, [4]> var_9911_begin_0 = const()[name = tensor<string, []>("op_9911_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 768])];
            tensor<int32, [4]> var_9911_end_0 = const()[name = tensor<string, []>("op_9911_end_0"), val = tensor<int32, [4]>([2, 77, 1, 832])];
            tensor<bool, [4]> var_9911_end_mask_0 = const()[name = tensor<string, []>("op_9911_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9911_cast = slice_by_index(begin = var_9911_begin_0, end = var_9911_end_0, end_mask = var_9911_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9911_cast")];
            tensor<int32, [4]> var_9915_begin_0 = const()[name = tensor<string, []>("op_9915_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 832])];
            tensor<int32, [4]> var_9915_end_0 = const()[name = tensor<string, []>("op_9915_end_0"), val = tensor<int32, [4]>([2, 77, 1, 896])];
            tensor<bool, [4]> var_9915_end_mask_0 = const()[name = tensor<string, []>("op_9915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9915_cast = slice_by_index(begin = var_9915_begin_0, end = var_9915_end_0, end_mask = var_9915_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9915_cast")];
            tensor<int32, [4]> var_9919_begin_0 = const()[name = tensor<string, []>("op_9919_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 896])];
            tensor<int32, [4]> var_9919_end_0 = const()[name = tensor<string, []>("op_9919_end_0"), val = tensor<int32, [4]>([2, 77, 1, 960])];
            tensor<bool, [4]> var_9919_end_mask_0 = const()[name = tensor<string, []>("op_9919_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9919_cast = slice_by_index(begin = var_9919_begin_0, end = var_9919_end_0, end_mask = var_9919_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9919_cast")];
            tensor<int32, [4]> var_9923_begin_0 = const()[name = tensor<string, []>("op_9923_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 960])];
            tensor<int32, [4]> var_9923_end_0 = const()[name = tensor<string, []>("op_9923_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1024])];
            tensor<bool, [4]> var_9923_end_mask_0 = const()[name = tensor<string, []>("op_9923_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9923_cast = slice_by_index(begin = var_9923_begin_0, end = var_9923_end_0, end_mask = var_9923_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9923_cast")];
            tensor<int32, [4]> var_9927_begin_0 = const()[name = tensor<string, []>("op_9927_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_9927_end_0 = const()[name = tensor<string, []>("op_9927_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1088])];
            tensor<bool, [4]> var_9927_end_mask_0 = const()[name = tensor<string, []>("op_9927_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9927_cast = slice_by_index(begin = var_9927_begin_0, end = var_9927_end_0, end_mask = var_9927_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9927_cast")];
            tensor<int32, [4]> var_9931_begin_0 = const()[name = tensor<string, []>("op_9931_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1088])];
            tensor<int32, [4]> var_9931_end_0 = const()[name = tensor<string, []>("op_9931_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1152])];
            tensor<bool, [4]> var_9931_end_mask_0 = const()[name = tensor<string, []>("op_9931_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9931_cast = slice_by_index(begin = var_9931_begin_0, end = var_9931_end_0, end_mask = var_9931_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9931_cast")];
            tensor<int32, [4]> var_9935_begin_0 = const()[name = tensor<string, []>("op_9935_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1152])];
            tensor<int32, [4]> var_9935_end_0 = const()[name = tensor<string, []>("op_9935_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1216])];
            tensor<bool, [4]> var_9935_end_mask_0 = const()[name = tensor<string, []>("op_9935_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9935_cast = slice_by_index(begin = var_9935_begin_0, end = var_9935_end_0, end_mask = var_9935_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9935_cast")];
            tensor<int32, [4]> var_9939_begin_0 = const()[name = tensor<string, []>("op_9939_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1216])];
            tensor<int32, [4]> var_9939_end_0 = const()[name = tensor<string, []>("op_9939_end_0"), val = tensor<int32, [4]>([2, 77, 1, 1280])];
            tensor<bool, [4]> var_9939_end_mask_0 = const()[name = tensor<string, []>("op_9939_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_9939_cast = slice_by_index(begin = var_9939_begin_0, end = var_9939_end_0, end_mask = var_9939_end_mask_0, x = transpose_12)[name = tensor<string, []>("op_9939_cast")];
            tensor<int32, [4]> var_9941_begin_0 = const()[name = tensor<string, []>("op_9941_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_9941_end_0 = const()[name = tensor<string, []>("op_9941_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_9941_end_mask_0 = const()[name = tensor<string, []>("op_9941_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9941_cast = slice_by_index(begin = var_9941_begin_0, end = var_9941_end_0, end_mask = var_9941_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9941_cast")];
            tensor<int32, [4]> var_9945_begin_0 = const()[name = tensor<string, []>("op_9945_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_9945_end_0 = const()[name = tensor<string, []>("op_9945_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_9945_end_mask_0 = const()[name = tensor<string, []>("op_9945_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9945_cast = slice_by_index(begin = var_9945_begin_0, end = var_9945_end_0, end_mask = var_9945_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9945_cast")];
            tensor<int32, [4]> var_9949_begin_0 = const()[name = tensor<string, []>("op_9949_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_9949_end_0 = const()[name = tensor<string, []>("op_9949_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_9949_end_mask_0 = const()[name = tensor<string, []>("op_9949_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9949_cast = slice_by_index(begin = var_9949_begin_0, end = var_9949_end_0, end_mask = var_9949_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9949_cast")];
            tensor<int32, [4]> var_9953_begin_0 = const()[name = tensor<string, []>("op_9953_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_9953_end_0 = const()[name = tensor<string, []>("op_9953_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_9953_end_mask_0 = const()[name = tensor<string, []>("op_9953_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9953_cast = slice_by_index(begin = var_9953_begin_0, end = var_9953_end_0, end_mask = var_9953_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9953_cast")];
            tensor<int32, [4]> var_9957_begin_0 = const()[name = tensor<string, []>("op_9957_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_9957_end_0 = const()[name = tensor<string, []>("op_9957_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_9957_end_mask_0 = const()[name = tensor<string, []>("op_9957_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9957_cast = slice_by_index(begin = var_9957_begin_0, end = var_9957_end_0, end_mask = var_9957_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9957_cast")];
            tensor<int32, [4]> var_9961_begin_0 = const()[name = tensor<string, []>("op_9961_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_9961_end_0 = const()[name = tensor<string, []>("op_9961_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_9961_end_mask_0 = const()[name = tensor<string, []>("op_9961_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9961_cast = slice_by_index(begin = var_9961_begin_0, end = var_9961_end_0, end_mask = var_9961_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9961_cast")];
            tensor<int32, [4]> var_9965_begin_0 = const()[name = tensor<string, []>("op_9965_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_9965_end_0 = const()[name = tensor<string, []>("op_9965_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_9965_end_mask_0 = const()[name = tensor<string, []>("op_9965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9965_cast = slice_by_index(begin = var_9965_begin_0, end = var_9965_end_0, end_mask = var_9965_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9965_cast")];
            tensor<int32, [4]> var_9969_begin_0 = const()[name = tensor<string, []>("op_9969_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_9969_end_0 = const()[name = tensor<string, []>("op_9969_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_9969_end_mask_0 = const()[name = tensor<string, []>("op_9969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9969_cast = slice_by_index(begin = var_9969_begin_0, end = var_9969_end_0, end_mask = var_9969_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9969_cast")];
            tensor<int32, [4]> var_9973_begin_0 = const()[name = tensor<string, []>("op_9973_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_9973_end_0 = const()[name = tensor<string, []>("op_9973_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_9973_end_mask_0 = const()[name = tensor<string, []>("op_9973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9973_cast = slice_by_index(begin = var_9973_begin_0, end = var_9973_end_0, end_mask = var_9973_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9973_cast")];
            tensor<int32, [4]> var_9977_begin_0 = const()[name = tensor<string, []>("op_9977_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_9977_end_0 = const()[name = tensor<string, []>("op_9977_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_9977_end_mask_0 = const()[name = tensor<string, []>("op_9977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9977_cast = slice_by_index(begin = var_9977_begin_0, end = var_9977_end_0, end_mask = var_9977_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9977_cast")];
            tensor<int32, [4]> var_9981_begin_0 = const()[name = tensor<string, []>("op_9981_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_9981_end_0 = const()[name = tensor<string, []>("op_9981_end_0"), val = tensor<int32, [4]>([2, 704, 1, 77])];
            tensor<bool, [4]> var_9981_end_mask_0 = const()[name = tensor<string, []>("op_9981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9981_cast = slice_by_index(begin = var_9981_begin_0, end = var_9981_end_0, end_mask = var_9981_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9981_cast")];
            tensor<int32, [4]> var_9985_begin_0 = const()[name = tensor<string, []>("op_9985_begin_0"), val = tensor<int32, [4]>([0, 704, 0, 0])];
            tensor<int32, [4]> var_9985_end_0 = const()[name = tensor<string, []>("op_9985_end_0"), val = tensor<int32, [4]>([2, 768, 1, 77])];
            tensor<bool, [4]> var_9985_end_mask_0 = const()[name = tensor<string, []>("op_9985_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9985_cast = slice_by_index(begin = var_9985_begin_0, end = var_9985_end_0, end_mask = var_9985_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9985_cast")];
            tensor<int32, [4]> var_9989_begin_0 = const()[name = tensor<string, []>("op_9989_begin_0"), val = tensor<int32, [4]>([0, 768, 0, 0])];
            tensor<int32, [4]> var_9989_end_0 = const()[name = tensor<string, []>("op_9989_end_0"), val = tensor<int32, [4]>([2, 832, 1, 77])];
            tensor<bool, [4]> var_9989_end_mask_0 = const()[name = tensor<string, []>("op_9989_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9989_cast = slice_by_index(begin = var_9989_begin_0, end = var_9989_end_0, end_mask = var_9989_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9989_cast")];
            tensor<int32, [4]> var_9993_begin_0 = const()[name = tensor<string, []>("op_9993_begin_0"), val = tensor<int32, [4]>([0, 832, 0, 0])];
            tensor<int32, [4]> var_9993_end_0 = const()[name = tensor<string, []>("op_9993_end_0"), val = tensor<int32, [4]>([2, 896, 1, 77])];
            tensor<bool, [4]> var_9993_end_mask_0 = const()[name = tensor<string, []>("op_9993_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9993_cast = slice_by_index(begin = var_9993_begin_0, end = var_9993_end_0, end_mask = var_9993_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9993_cast")];
            tensor<int32, [4]> var_9997_begin_0 = const()[name = tensor<string, []>("op_9997_begin_0"), val = tensor<int32, [4]>([0, 896, 0, 0])];
            tensor<int32, [4]> var_9997_end_0 = const()[name = tensor<string, []>("op_9997_end_0"), val = tensor<int32, [4]>([2, 960, 1, 77])];
            tensor<bool, [4]> var_9997_end_mask_0 = const()[name = tensor<string, []>("op_9997_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_9997_cast = slice_by_index(begin = var_9997_begin_0, end = var_9997_end_0, end_mask = var_9997_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_9997_cast")];
            tensor<int32, [4]> var_10001_begin_0 = const()[name = tensor<string, []>("op_10001_begin_0"), val = tensor<int32, [4]>([0, 960, 0, 0])];
            tensor<int32, [4]> var_10001_end_0 = const()[name = tensor<string, []>("op_10001_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 77])];
            tensor<bool, [4]> var_10001_end_mask_0 = const()[name = tensor<string, []>("op_10001_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10001_cast = slice_by_index(begin = var_10001_begin_0, end = var_10001_end_0, end_mask = var_10001_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_10001_cast")];
            tensor<int32, [4]> var_10005_begin_0 = const()[name = tensor<string, []>("op_10005_begin_0"), val = tensor<int32, [4]>([0, 1024, 0, 0])];
            tensor<int32, [4]> var_10005_end_0 = const()[name = tensor<string, []>("op_10005_end_0"), val = tensor<int32, [4]>([2, 1088, 1, 77])];
            tensor<bool, [4]> var_10005_end_mask_0 = const()[name = tensor<string, []>("op_10005_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10005_cast = slice_by_index(begin = var_10005_begin_0, end = var_10005_end_0, end_mask = var_10005_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_10005_cast")];
            tensor<int32, [4]> var_10009_begin_0 = const()[name = tensor<string, []>("op_10009_begin_0"), val = tensor<int32, [4]>([0, 1088, 0, 0])];
            tensor<int32, [4]> var_10009_end_0 = const()[name = tensor<string, []>("op_10009_end_0"), val = tensor<int32, [4]>([2, 1152, 1, 77])];
            tensor<bool, [4]> var_10009_end_mask_0 = const()[name = tensor<string, []>("op_10009_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10009_cast = slice_by_index(begin = var_10009_begin_0, end = var_10009_end_0, end_mask = var_10009_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_10009_cast")];
            tensor<int32, [4]> var_10013_begin_0 = const()[name = tensor<string, []>("op_10013_begin_0"), val = tensor<int32, [4]>([0, 1152, 0, 0])];
            tensor<int32, [4]> var_10013_end_0 = const()[name = tensor<string, []>("op_10013_end_0"), val = tensor<int32, [4]>([2, 1216, 1, 77])];
            tensor<bool, [4]> var_10013_end_mask_0 = const()[name = tensor<string, []>("op_10013_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10013_cast = slice_by_index(begin = var_10013_begin_0, end = var_10013_end_0, end_mask = var_10013_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_10013_cast")];
            tensor<int32, [4]> var_10017_begin_0 = const()[name = tensor<string, []>("op_10017_begin_0"), val = tensor<int32, [4]>([0, 1216, 0, 0])];
            tensor<int32, [4]> var_10017_end_0 = const()[name = tensor<string, []>("op_10017_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 77])];
            tensor<bool, [4]> var_10017_end_mask_0 = const()[name = tensor<string, []>("op_10017_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10017_cast = slice_by_index(begin = var_10017_begin_0, end = var_10017_end_0, end_mask = var_10017_end_mask_0, x = v_39_cast)[name = tensor<string, []>("op_10017_cast")];
            tensor<string, []> var_10021_equation_0 = const()[name = tensor<string, []>("op_10021_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10021_cast = einsum(equation = var_10021_equation_0, values = (var_9863_cast, var_9780_cast))[name = tensor<string, []>("op_10021_cast")];
            tensor<fp16, []> var_10022_to_fp16 = const()[name = tensor<string, []>("op_10022_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_441_cast = mul(x = var_10021_cast, y = var_10022_to_fp16)[name = tensor<string, []>("aw_441_cast")];
            tensor<string, []> var_10025_equation_0 = const()[name = tensor<string, []>("op_10025_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10025_cast = einsum(equation = var_10025_equation_0, values = (var_9867_cast, var_9784_cast))[name = tensor<string, []>("op_10025_cast")];
            tensor<fp16, []> var_10026_to_fp16 = const()[name = tensor<string, []>("op_10026_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_443_cast = mul(x = var_10025_cast, y = var_10026_to_fp16)[name = tensor<string, []>("aw_443_cast")];
            tensor<string, []> var_10029_equation_0 = const()[name = tensor<string, []>("op_10029_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10029_cast = einsum(equation = var_10029_equation_0, values = (var_9871_cast, var_9788_cast))[name = tensor<string, []>("op_10029_cast")];
            tensor<fp16, []> var_10030_to_fp16 = const()[name = tensor<string, []>("op_10030_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_445_cast = mul(x = var_10029_cast, y = var_10030_to_fp16)[name = tensor<string, []>("aw_445_cast")];
            tensor<string, []> var_10033_equation_0 = const()[name = tensor<string, []>("op_10033_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10033_cast = einsum(equation = var_10033_equation_0, values = (var_9875_cast, var_9792_cast))[name = tensor<string, []>("op_10033_cast")];
            tensor<fp16, []> var_10034_to_fp16 = const()[name = tensor<string, []>("op_10034_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_447_cast = mul(x = var_10033_cast, y = var_10034_to_fp16)[name = tensor<string, []>("aw_447_cast")];
            tensor<string, []> var_10037_equation_0 = const()[name = tensor<string, []>("op_10037_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10037_cast = einsum(equation = var_10037_equation_0, values = (var_9879_cast, var_9796_cast))[name = tensor<string, []>("op_10037_cast")];
            tensor<fp16, []> var_10038_to_fp16 = const()[name = tensor<string, []>("op_10038_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_449_cast = mul(x = var_10037_cast, y = var_10038_to_fp16)[name = tensor<string, []>("aw_449_cast")];
            tensor<string, []> var_10041_equation_0 = const()[name = tensor<string, []>("op_10041_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10041_cast = einsum(equation = var_10041_equation_0, values = (var_9883_cast, var_9800_cast))[name = tensor<string, []>("op_10041_cast")];
            tensor<fp16, []> var_10042_to_fp16 = const()[name = tensor<string, []>("op_10042_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_451_cast = mul(x = var_10041_cast, y = var_10042_to_fp16)[name = tensor<string, []>("aw_451_cast")];
            tensor<string, []> var_10045_equation_0 = const()[name = tensor<string, []>("op_10045_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10045_cast = einsum(equation = var_10045_equation_0, values = (var_9887_cast, var_9804_cast))[name = tensor<string, []>("op_10045_cast")];
            tensor<fp16, []> var_10046_to_fp16 = const()[name = tensor<string, []>("op_10046_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_453_cast = mul(x = var_10045_cast, y = var_10046_to_fp16)[name = tensor<string, []>("aw_453_cast")];
            tensor<string, []> var_10049_equation_0 = const()[name = tensor<string, []>("op_10049_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10049_cast = einsum(equation = var_10049_equation_0, values = (var_9891_cast, var_9808_cast))[name = tensor<string, []>("op_10049_cast")];
            tensor<fp16, []> var_10050_to_fp16 = const()[name = tensor<string, []>("op_10050_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_455_cast = mul(x = var_10049_cast, y = var_10050_to_fp16)[name = tensor<string, []>("aw_455_cast")];
            tensor<string, []> var_10053_equation_0 = const()[name = tensor<string, []>("op_10053_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10053_cast = einsum(equation = var_10053_equation_0, values = (var_9895_cast, var_9812_cast))[name = tensor<string, []>("op_10053_cast")];
            tensor<fp16, []> var_10054_to_fp16 = const()[name = tensor<string, []>("op_10054_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_457_cast = mul(x = var_10053_cast, y = var_10054_to_fp16)[name = tensor<string, []>("aw_457_cast")];
            tensor<string, []> var_10057_equation_0 = const()[name = tensor<string, []>("op_10057_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10057_cast = einsum(equation = var_10057_equation_0, values = (var_9899_cast, var_9816_cast))[name = tensor<string, []>("op_10057_cast")];
            tensor<fp16, []> var_10058_to_fp16 = const()[name = tensor<string, []>("op_10058_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_459_cast = mul(x = var_10057_cast, y = var_10058_to_fp16)[name = tensor<string, []>("aw_459_cast")];
            tensor<string, []> var_10061_equation_0 = const()[name = tensor<string, []>("op_10061_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10061_cast = einsum(equation = var_10061_equation_0, values = (var_9903_cast, var_9820_cast))[name = tensor<string, []>("op_10061_cast")];
            tensor<fp16, []> var_10062_to_fp16 = const()[name = tensor<string, []>("op_10062_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_461_cast = mul(x = var_10061_cast, y = var_10062_to_fp16)[name = tensor<string, []>("aw_461_cast")];
            tensor<string, []> var_10065_equation_0 = const()[name = tensor<string, []>("op_10065_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10065_cast = einsum(equation = var_10065_equation_0, values = (var_9907_cast, var_9824_cast))[name = tensor<string, []>("op_10065_cast")];
            tensor<fp16, []> var_10066_to_fp16 = const()[name = tensor<string, []>("op_10066_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_463_cast = mul(x = var_10065_cast, y = var_10066_to_fp16)[name = tensor<string, []>("aw_463_cast")];
            tensor<string, []> var_10069_equation_0 = const()[name = tensor<string, []>("op_10069_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10069_cast = einsum(equation = var_10069_equation_0, values = (var_9911_cast, var_9828_cast))[name = tensor<string, []>("op_10069_cast")];
            tensor<fp16, []> var_10070_to_fp16 = const()[name = tensor<string, []>("op_10070_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_465_cast = mul(x = var_10069_cast, y = var_10070_to_fp16)[name = tensor<string, []>("aw_465_cast")];
            tensor<string, []> var_10073_equation_0 = const()[name = tensor<string, []>("op_10073_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10073_cast = einsum(equation = var_10073_equation_0, values = (var_9915_cast, var_9832_cast))[name = tensor<string, []>("op_10073_cast")];
            tensor<fp16, []> var_10074_to_fp16 = const()[name = tensor<string, []>("op_10074_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_467_cast = mul(x = var_10073_cast, y = var_10074_to_fp16)[name = tensor<string, []>("aw_467_cast")];
            tensor<string, []> var_10077_equation_0 = const()[name = tensor<string, []>("op_10077_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10077_cast = einsum(equation = var_10077_equation_0, values = (var_9919_cast, var_9836_cast))[name = tensor<string, []>("op_10077_cast")];
            tensor<fp16, []> var_10078_to_fp16 = const()[name = tensor<string, []>("op_10078_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_469_cast = mul(x = var_10077_cast, y = var_10078_to_fp16)[name = tensor<string, []>("aw_469_cast")];
            tensor<string, []> var_10081_equation_0 = const()[name = tensor<string, []>("op_10081_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10081_cast = einsum(equation = var_10081_equation_0, values = (var_9923_cast, var_9840_cast))[name = tensor<string, []>("op_10081_cast")];
            tensor<fp16, []> var_10082_to_fp16 = const()[name = tensor<string, []>("op_10082_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_471_cast = mul(x = var_10081_cast, y = var_10082_to_fp16)[name = tensor<string, []>("aw_471_cast")];
            tensor<string, []> var_10085_equation_0 = const()[name = tensor<string, []>("op_10085_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10085_cast = einsum(equation = var_10085_equation_0, values = (var_9927_cast, var_9844_cast))[name = tensor<string, []>("op_10085_cast")];
            tensor<fp16, []> var_10086_to_fp16 = const()[name = tensor<string, []>("op_10086_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_473_cast = mul(x = var_10085_cast, y = var_10086_to_fp16)[name = tensor<string, []>("aw_473_cast")];
            tensor<string, []> var_10089_equation_0 = const()[name = tensor<string, []>("op_10089_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10089_cast = einsum(equation = var_10089_equation_0, values = (var_9931_cast, var_9848_cast))[name = tensor<string, []>("op_10089_cast")];
            tensor<fp16, []> var_10090_to_fp16 = const()[name = tensor<string, []>("op_10090_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_475_cast = mul(x = var_10089_cast, y = var_10090_to_fp16)[name = tensor<string, []>("aw_475_cast")];
            tensor<string, []> var_10093_equation_0 = const()[name = tensor<string, []>("op_10093_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10093_cast = einsum(equation = var_10093_equation_0, values = (var_9935_cast, var_9852_cast))[name = tensor<string, []>("op_10093_cast")];
            tensor<fp16, []> var_10094_to_fp16 = const()[name = tensor<string, []>("op_10094_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_477_cast = mul(x = var_10093_cast, y = var_10094_to_fp16)[name = tensor<string, []>("aw_477_cast")];
            tensor<string, []> var_10097_equation_0 = const()[name = tensor<string, []>("op_10097_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 256]> var_10097_cast = einsum(equation = var_10097_equation_0, values = (var_9939_cast, var_9856_cast))[name = tensor<string, []>("op_10097_cast")];
            tensor<fp16, []> var_10098_to_fp16 = const()[name = tensor<string, []>("op_10098_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 256]> aw_cast = mul(x = var_10097_cast, y = var_10098_to_fp16)[name = tensor<string, []>("aw_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10100_cast = softmax(axis = var_7228, x = aw_441_cast)[name = tensor<string, []>("op_10100_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10101_cast = softmax(axis = var_7228, x = aw_443_cast)[name = tensor<string, []>("op_10101_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10102_cast = softmax(axis = var_7228, x = aw_445_cast)[name = tensor<string, []>("op_10102_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10103_cast = softmax(axis = var_7228, x = aw_447_cast)[name = tensor<string, []>("op_10103_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10104_cast = softmax(axis = var_7228, x = aw_449_cast)[name = tensor<string, []>("op_10104_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10105_cast = softmax(axis = var_7228, x = aw_451_cast)[name = tensor<string, []>("op_10105_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10106_cast = softmax(axis = var_7228, x = aw_453_cast)[name = tensor<string, []>("op_10106_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10107_cast = softmax(axis = var_7228, x = aw_455_cast)[name = tensor<string, []>("op_10107_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10108_cast = softmax(axis = var_7228, x = aw_457_cast)[name = tensor<string, []>("op_10108_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10109_cast = softmax(axis = var_7228, x = aw_459_cast)[name = tensor<string, []>("op_10109_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10110_cast = softmax(axis = var_7228, x = aw_461_cast)[name = tensor<string, []>("op_10110_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10111_cast = softmax(axis = var_7228, x = aw_463_cast)[name = tensor<string, []>("op_10111_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10112_cast = softmax(axis = var_7228, x = aw_465_cast)[name = tensor<string, []>("op_10112_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10113_cast = softmax(axis = var_7228, x = aw_467_cast)[name = tensor<string, []>("op_10113_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10114_cast = softmax(axis = var_7228, x = aw_469_cast)[name = tensor<string, []>("op_10114_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10115_cast = softmax(axis = var_7228, x = aw_471_cast)[name = tensor<string, []>("op_10115_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10116_cast = softmax(axis = var_7228, x = aw_473_cast)[name = tensor<string, []>("op_10116_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10117_cast = softmax(axis = var_7228, x = aw_475_cast)[name = tensor<string, []>("op_10117_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10118_cast = softmax(axis = var_7228, x = aw_477_cast)[name = tensor<string, []>("op_10118_cast")];
            tensor<fp16, [2, 77, 1, 256]> var_10119_cast = softmax(axis = var_7228, x = aw_cast)[name = tensor<string, []>("op_10119_cast")];
            tensor<string, []> var_10121_equation_0 = const()[name = tensor<string, []>("op_10121_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10121_cast = einsum(equation = var_10121_equation_0, values = (var_9941_cast, var_10100_cast))[name = tensor<string, []>("op_10121_cast")];
            tensor<string, []> var_10123_equation_0 = const()[name = tensor<string, []>("op_10123_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10123_cast = einsum(equation = var_10123_equation_0, values = (var_9945_cast, var_10101_cast))[name = tensor<string, []>("op_10123_cast")];
            tensor<string, []> var_10125_equation_0 = const()[name = tensor<string, []>("op_10125_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10125_cast = einsum(equation = var_10125_equation_0, values = (var_9949_cast, var_10102_cast))[name = tensor<string, []>("op_10125_cast")];
            tensor<string, []> var_10127_equation_0 = const()[name = tensor<string, []>("op_10127_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10127_cast = einsum(equation = var_10127_equation_0, values = (var_9953_cast, var_10103_cast))[name = tensor<string, []>("op_10127_cast")];
            tensor<string, []> var_10129_equation_0 = const()[name = tensor<string, []>("op_10129_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10129_cast = einsum(equation = var_10129_equation_0, values = (var_9957_cast, var_10104_cast))[name = tensor<string, []>("op_10129_cast")];
            tensor<string, []> var_10131_equation_0 = const()[name = tensor<string, []>("op_10131_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10131_cast = einsum(equation = var_10131_equation_0, values = (var_9961_cast, var_10105_cast))[name = tensor<string, []>("op_10131_cast")];
            tensor<string, []> var_10133_equation_0 = const()[name = tensor<string, []>("op_10133_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10133_cast = einsum(equation = var_10133_equation_0, values = (var_9965_cast, var_10106_cast))[name = tensor<string, []>("op_10133_cast")];
            tensor<string, []> var_10135_equation_0 = const()[name = tensor<string, []>("op_10135_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10135_cast = einsum(equation = var_10135_equation_0, values = (var_9969_cast, var_10107_cast))[name = tensor<string, []>("op_10135_cast")];
            tensor<string, []> var_10137_equation_0 = const()[name = tensor<string, []>("op_10137_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10137_cast = einsum(equation = var_10137_equation_0, values = (var_9973_cast, var_10108_cast))[name = tensor<string, []>("op_10137_cast")];
            tensor<string, []> var_10139_equation_0 = const()[name = tensor<string, []>("op_10139_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10139_cast = einsum(equation = var_10139_equation_0, values = (var_9977_cast, var_10109_cast))[name = tensor<string, []>("op_10139_cast")];
            tensor<string, []> var_10141_equation_0 = const()[name = tensor<string, []>("op_10141_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10141_cast = einsum(equation = var_10141_equation_0, values = (var_9981_cast, var_10110_cast))[name = tensor<string, []>("op_10141_cast")];
            tensor<string, []> var_10143_equation_0 = const()[name = tensor<string, []>("op_10143_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10143_cast = einsum(equation = var_10143_equation_0, values = (var_9985_cast, var_10111_cast))[name = tensor<string, []>("op_10143_cast")];
            tensor<string, []> var_10145_equation_0 = const()[name = tensor<string, []>("op_10145_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10145_cast = einsum(equation = var_10145_equation_0, values = (var_9989_cast, var_10112_cast))[name = tensor<string, []>("op_10145_cast")];
            tensor<string, []> var_10147_equation_0 = const()[name = tensor<string, []>("op_10147_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10147_cast = einsum(equation = var_10147_equation_0, values = (var_9993_cast, var_10113_cast))[name = tensor<string, []>("op_10147_cast")];
            tensor<string, []> var_10149_equation_0 = const()[name = tensor<string, []>("op_10149_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10149_cast = einsum(equation = var_10149_equation_0, values = (var_9997_cast, var_10114_cast))[name = tensor<string, []>("op_10149_cast")];
            tensor<string, []> var_10151_equation_0 = const()[name = tensor<string, []>("op_10151_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10151_cast = einsum(equation = var_10151_equation_0, values = (var_10001_cast, var_10115_cast))[name = tensor<string, []>("op_10151_cast")];
            tensor<string, []> var_10153_equation_0 = const()[name = tensor<string, []>("op_10153_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10153_cast = einsum(equation = var_10153_equation_0, values = (var_10005_cast, var_10116_cast))[name = tensor<string, []>("op_10153_cast")];
            tensor<string, []> var_10155_equation_0 = const()[name = tensor<string, []>("op_10155_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10155_cast = einsum(equation = var_10155_equation_0, values = (var_10009_cast, var_10117_cast))[name = tensor<string, []>("op_10155_cast")];
            tensor<string, []> var_10157_equation_0 = const()[name = tensor<string, []>("op_10157_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10157_cast = einsum(equation = var_10157_equation_0, values = (var_10013_cast, var_10118_cast))[name = tensor<string, []>("op_10157_cast")];
            tensor<string, []> var_10159_equation_0 = const()[name = tensor<string, []>("op_10159_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 256]> var_10159_cast = einsum(equation = var_10159_equation_0, values = (var_10017_cast, var_10119_cast))[name = tensor<string, []>("op_10159_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_7228, interleave = input_355_interleave_0, values = (var_10121_cast, var_10123_cast, var_10125_cast, var_10127_cast, var_10129_cast, var_10131_cast, var_10133_cast, var_10135_cast, var_10137_cast, var_10139_cast, var_10141_cast, var_10143_cast, var_10145_cast, var_10147_cast, var_10149_cast, var_10151_cast, var_10153_cast, var_10155_cast, var_10157_cast, var_10159_cast))[name = tensor<string, []>("input_355_cast")];
            tensor<int32, [2]> var_10165 = const()[name = tensor<string, []>("op_10165"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10167 = const()[name = tensor<string, []>("op_10167"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10169_pad_type_0 = const()[name = tensor<string, []>("op_10169_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10169_pad_0 = const()[name = tensor<string, []>("op_10169_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184595328))), lut = tensor<fp16, [4]>([-0x1.9ccp-5, -0x1.eep-7, 0x1.e28p-7, 0x1.9a8p-5]), 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, []>(185004992)))];
            tensor<fp16, [2, 1280, 1, 256]> var_10169_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_10167, groups = var_7228, pad = var_10169_pad_0, pad_type = var_10169_pad_type_0, strides = var_10165, 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_10169_cast")];
            tensor<fp16, [2, 1280, 1, 256]> inputs_59_cast = add(x = var_10169_cast, y = inputs_57_cast)[name = tensor<string, []>("inputs_59_cast")];
            tensor<int32, [1]> var_10173 = const()[name = tensor<string, []>("op_10173"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> channels_mean_59_cast = reduce_mean(axes = var_10173, keep_dims = var_7223, 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_10177 = const()[name = tensor<string, []>("op_10177"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 256]> var_10178_cast = reduce_mean(axes = var_10177, keep_dims = var_7223, x = zero_mean_sq_59_cast)[name = tensor<string, []>("op_10178_cast")];
            tensor<fp16, []> var_10179_to_fp16 = const()[name = tensor<string, []>("op_10179_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 256]> var_10180_cast = add(x = var_10178_cast, y = var_10179_to_fp16)[name = tensor<string, []>("op_10180_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_10180_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_10184_to_fp16 = const()[name = tensor<string, []>("op_10184_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185007616)))];
            tensor<fp16, [2, 1280, 1, 256]> var_10185_cast = add(x = out_59_cast, y = var_10184_to_fp16)[name = tensor<string, []>("op_10185_cast")];
            tensor<fp16, [1, 1280, 1, 1]> var_10187_to_fp16 = const()[name = tensor<string, []>("op_10187_to_fp16"), val = tensor<fp16, [1, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185010240)))];
            tensor<fp16, [2, 1280, 1, 256]> input_357_cast = mul(x = var_10185_cast, y = var_10187_to_fp16)[name = tensor<string, []>("input_357_cast")];
            tensor<int32, [2]> var_10195 = const()[name = tensor<string, []>("op_10195"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10197 = const()[name = tensor<string, []>("op_10197"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10199_pad_type_0 = const()[name = tensor<string, []>("op_10199_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10199_pad_0 = const()[name = tensor<string, []>("op_10199_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, [3276800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185012864))), lut = tensor<fp16, [4]>([-0x1.6cp-4, -0x1.af8p-6, 0x1.b3p-6, 0x1.6dp-4]), 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, [2560]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188289728))), lut = tensor<fp16, [4]>([-0x1.638p-3, -0x1.34cp-4, -0x1.16p-6, 0x1.ac4p-6]), 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_10199_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_10197, groups = var_7228, pad = var_10199_pad_0, pad_type = var_10199_pad_type_0, strides = var_10195, 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_10199_cast")];
            tensor<int32, [2]> var_10200_split_sizes_0 = const()[name = tensor<string, []>("op_10200_split_sizes_0"), val = tensor<int32, [2]>([5120, 5120])];
            tensor<int32, []> var_10200_axis_0 = const()[name = tensor<string, []>("op_10200_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 5120, 1, 256]> var_10200_cast_0, tensor<fp16, [2, 5120, 1, 256]> var_10200_cast_1 = split(axis = var_10200_axis_0, split_sizes = var_10200_split_sizes_0, x = var_10199_cast)[name = tensor<string, []>("op_10200_cast")];
            tensor<string, []> var_10202_mode_0 = const()[name = tensor<string, []>("op_10202_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 5120, 1, 256]> var_10202_cast = gelu(mode = var_10202_mode_0, x = var_10200_cast_1)[name = tensor<string, []>("op_10202_cast")];
            tensor<fp16, [2, 5120, 1, 256]> input_359_cast = mul(x = var_10200_cast_0, y = var_10202_cast)[name = tensor<string, []>("input_359_cast")];
            tensor<int32, [2]> var_10206 = const()[name = tensor<string, []>("op_10206"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10208 = const()[name = tensor<string, []>("op_10208"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10210_pad_type_0 = const()[name = tensor<string, []>("op_10210_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10210_pad_0 = const()[name = tensor<string, []>("op_10210_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, [1638400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188292352))), lut = tensor<fp16, [4]>([-0x1.5e8p-4, -0x1.a38p-6, 0x1.a1p-6, 0x1.5dcp-4]), 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, []>(189930816)))];
            tensor<fp16, [2, 1280, 1, 256]> var_10210_cast = conv(bias = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_10208, groups = var_7228, pad = var_10210_pad_0, pad_type = var_10210_pad_type_0, strides = var_10206, weight = up_blocks_1_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_359_cast)[name = tensor<string, []>("op_10210_cast")];
            tensor<fp16, [2, 1280, 1, 256]> hidden_states_213_cast = add(x = var_10210_cast, y = inputs_59_cast)[name = tensor<string, []>("hidden_states_213_cast")];
            tensor<int32, [4]> var_10212 = const()[name = tensor<string, []>("op_10212"), val = tensor<int32, [4]>([2, 1280, 16, 16])];
            tensor<fp16, [2, 1280, 16, 16]> input_361_cast = reshape(shape = var_10212, x = hidden_states_213_cast)[name = tensor<string, []>("input_361_cast")];
            tensor<int32, [2]> var_10216 = const()[name = tensor<string, []>("op_10216"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10218 = const()[name = tensor<string, []>("op_10218"), 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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189933440))), lut = tensor<fp16, [4]>([-0x1.4bcp-4, -0x1.91cp-6, 0x1.8e8p-6, 0x1.4bp-4]), 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, []>(190343104)))];
            tensor<fp16, [2, 1280, 16, 16]> hidden_states_215_cast = conv(bias = up_blocks_1_attentions_2_proj_out_bias_to_fp16, dilations = var_10218, groups = var_7228, pad = hidden_states_215_pad_0, pad_type = hidden_states_215_pad_type_0, strides = var_10216, 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_10227 = const()[name = tensor<string, []>("op_10227"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10229 = const()[name = tensor<string, []>("op_10229"), 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, [3686400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190345728))), lut = tensor<fp16, [4]>([-0x1.504p-5, -0x1.81p-7, 0x1.82p-7, 0x1.50cp-5]), 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, []>(194032192)))];
            tensor<fp16, [2, 1280, 32, 32]> hidden_states_217_cast = conv(bias = up_blocks_1_upsamplers_0_conv_bias_to_fp16, dilations = var_10229, groups = var_7228, pad = hidden_states_217_pad_0, pad_type = hidden_states_217_pad_type_0, strides = var_10227, 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_10234 = const()[name = tensor<string, []>("op_10234"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_10252 = const()[name = tensor<string, []>("op_10252"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_10257 = const()[name = tensor<string, []>("op_10257"), 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_10257, 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, []>(194034816)))];
            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, []>(194038720)))];
            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_10286 = const()[name = tensor<string, []>("op_10286"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10288 = const()[name = tensor<string, []>("op_10288"), 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, [2764800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194042624))), lut = tensor<fp16, [4]>([-0x1.8f8p-4, -0x1.a9cp-6, 0x1.75cp-6, 0x1.79p-4]), 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, []>(196807488)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_219_cast = conv(bias = up_blocks_2_resnets_0_conv1_bias_to_fp16, dilations = var_10288, groups = var_10257, pad = hidden_states_219_pad_0, pad_type = hidden_states_219_pad_type_0, strides = var_10286, 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_10294 = const()[name = tensor<string, []>("op_10294"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10296 = const()[name = tensor<string, []>("op_10296"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196808832))), lut = tensor<fp16, [4]>([-0x1.04cp-4, -0x1.c7p-9, 0x1.d4p-9, 0x1.32p-4]), 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, []>(197013696)))];
            tensor<fp16, [2, 640, 1, 1]> temb_33_cast = conv(bias = up_blocks_2_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_10296, groups = var_10257, pad = temb_33_pad_0, pad_type = temb_33_pad_type_0, strides = var_10294, 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, []>(197015040)))];
            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, []>(197016384)))];
            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_10306 = const()[name = tensor<string, []>("op_10306"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10308 = const()[name = tensor<string, []>("op_10308"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197017728))), lut = tensor<fp16, [4]>([-0x1.1ecp-3, -0x1.1acp-5, 0x1.214p-5, 0x1.24p-3]), 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, []>(197939392)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_221_cast = conv(bias = up_blocks_2_resnets_0_conv2_bias_to_fp16, dilations = var_10308, groups = var_10257, pad = hidden_states_221_pad_0, pad_type = hidden_states_221_pad_type_0, strides = var_10306, 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_10313 = const()[name = tensor<string, []>("op_10313"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10315 = const()[name = tensor<string, []>("op_10315"), 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, [307200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197940736))), lut = tensor<fp16, [4]>([-0x1.5dp-4, -0x1.c6p-10, 0x1.864p-5, 0x1.de8p-4]), 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, []>(198248000)))];
            tensor<fp16, [2, 640, 32, 32]> x_17_cast = conv(bias = up_blocks_2_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_10315, groups = var_10257, pad = x_17_pad_0, pad_type = x_17_pad_type_0, strides = var_10313, 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, []>(198249344)))];
            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, []>(198250688)))];
            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_10335 = const()[name = tensor<string, []>("op_10335"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10337 = const()[name = tensor<string, []>("op_10337"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198252032))), lut = tensor<fp16, [4]>([-0x1.d7p-4, -0x1.1acp-5, 0x1.1bp-5, 0x1.d8cp-4]), 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, []>(198354496)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_225_cast = conv(bias = up_blocks_2_attentions_0_proj_in_bias_to_fp16, dilations = var_10337, groups = var_10257, pad = hidden_states_225_pad_0, pad_type = hidden_states_225_pad_type_0, strides = var_10335, 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_10342 = const()[name = tensor<string, []>("op_10342"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_61_cast = reshape(shape = var_10342, x = hidden_states_225_cast)[name = tensor<string, []>("inputs_61_cast")];
            tensor<int32, [1]> var_10352 = const()[name = tensor<string, []>("op_10352"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_61_cast = reduce_mean(axes = var_10352, keep_dims = var_10252, 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_10356 = const()[name = tensor<string, []>("op_10356"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10357_cast = reduce_mean(axes = var_10356, keep_dims = var_10252, x = zero_mean_sq_61_cast)[name = tensor<string, []>("op_10357_cast")];
            tensor<fp16, []> var_10358_to_fp16 = const()[name = tensor<string, []>("op_10358_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10359_cast = add(x = var_10357_cast, y = var_10358_to_fp16)[name = tensor<string, []>("op_10359_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_10359_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_10363_to_fp16 = const()[name = tensor<string, []>("op_10363_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198355840)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10364_cast = add(x = out_61_cast, y = var_10363_to_fp16)[name = tensor<string, []>("op_10364_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_10366_to_fp16 = const()[name = tensor<string, []>("op_10366_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198357184)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_227_cast = mul(x = var_10364_cast, y = var_10366_to_fp16)[name = tensor<string, []>("hidden_states_227_cast")];
            tensor<int32, [2]> var_10373 = const()[name = tensor<string, []>("op_10373"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10375 = const()[name = tensor<string, []>("op_10375"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198358528))), lut = tensor<fp16, [4]>([-0x1.bd8p-4, -0x1.08cp-5, 0x1.04cp-5, 0x1.bdp-4]), 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_10375, groups = var_10257, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = var_10373, 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_10379 = const()[name = tensor<string, []>("op_10379"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10381 = const()[name = tensor<string, []>("op_10381"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198460992))), lut = tensor<fp16, [4]>([-0x1.b8p-4, -0x1.018p-5, 0x1.04p-5, 0x1.b94p-4]), 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_10381, groups = var_10257, pad = k_81_pad_0, pad_type = k_81_pad_type_0, strides = var_10379, 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_10385 = const()[name = tensor<string, []>("op_10385"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10387 = const()[name = tensor<string, []>("op_10387"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198563456))), lut = tensor<fp16, [4]>([-0x1.aep-4, -0x1.f8cp-6, 0x1.008p-5, 0x1.aep-4]), 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_10387, groups = var_10257, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = var_10385, 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_10391_begin_0 = const()[name = tensor<string, []>("op_10391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10391_end_0 = const()[name = tensor<string, []>("op_10391_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10391_end_mask_0 = const()[name = tensor<string, []>("op_10391_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10391_cast = slice_by_index(begin = var_10391_begin_0, end = var_10391_end_0, end_mask = var_10391_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10391_cast")];
            tensor<int32, [4]> var_10395_begin_0 = const()[name = tensor<string, []>("op_10395_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10395_end_0 = const()[name = tensor<string, []>("op_10395_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10395_end_mask_0 = const()[name = tensor<string, []>("op_10395_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10395_cast = slice_by_index(begin = var_10395_begin_0, end = var_10395_end_0, end_mask = var_10395_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10395_cast")];
            tensor<int32, [4]> var_10399_begin_0 = const()[name = tensor<string, []>("op_10399_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10399_end_0 = const()[name = tensor<string, []>("op_10399_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10399_end_mask_0 = const()[name = tensor<string, []>("op_10399_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10399_cast = slice_by_index(begin = var_10399_begin_0, end = var_10399_end_0, end_mask = var_10399_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10399_cast")];
            tensor<int32, [4]> var_10403_begin_0 = const()[name = tensor<string, []>("op_10403_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10403_end_0 = const()[name = tensor<string, []>("op_10403_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10403_end_mask_0 = const()[name = tensor<string, []>("op_10403_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10403_cast = slice_by_index(begin = var_10403_begin_0, end = var_10403_end_0, end_mask = var_10403_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10403_cast")];
            tensor<int32, [4]> var_10407_begin_0 = const()[name = tensor<string, []>("op_10407_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10407_end_0 = const()[name = tensor<string, []>("op_10407_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10407_end_mask_0 = const()[name = tensor<string, []>("op_10407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10407_cast = slice_by_index(begin = var_10407_begin_0, end = var_10407_end_0, end_mask = var_10407_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10407_cast")];
            tensor<int32, [4]> var_10411_begin_0 = const()[name = tensor<string, []>("op_10411_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10411_end_0 = const()[name = tensor<string, []>("op_10411_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10411_end_mask_0 = const()[name = tensor<string, []>("op_10411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10411_cast = slice_by_index(begin = var_10411_begin_0, end = var_10411_end_0, end_mask = var_10411_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10411_cast")];
            tensor<int32, [4]> var_10415_begin_0 = const()[name = tensor<string, []>("op_10415_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10415_end_0 = const()[name = tensor<string, []>("op_10415_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10415_end_mask_0 = const()[name = tensor<string, []>("op_10415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10415_cast = slice_by_index(begin = var_10415_begin_0, end = var_10415_end_0, end_mask = var_10415_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10415_cast")];
            tensor<int32, [4]> var_10419_begin_0 = const()[name = tensor<string, []>("op_10419_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10419_end_0 = const()[name = tensor<string, []>("op_10419_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10419_end_mask_0 = const()[name = tensor<string, []>("op_10419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10419_cast = slice_by_index(begin = var_10419_begin_0, end = var_10419_end_0, end_mask = var_10419_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10419_cast")];
            tensor<int32, [4]> var_10423_begin_0 = const()[name = tensor<string, []>("op_10423_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10423_end_0 = const()[name = tensor<string, []>("op_10423_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10423_end_mask_0 = const()[name = tensor<string, []>("op_10423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10423_cast = slice_by_index(begin = var_10423_begin_0, end = var_10423_end_0, end_mask = var_10423_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10423_cast")];
            tensor<int32, [4]> var_10427_begin_0 = const()[name = tensor<string, []>("op_10427_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10427_end_0 = const()[name = tensor<string, []>("op_10427_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10427_end_mask_0 = const()[name = tensor<string, []>("op_10427_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10427_cast = slice_by_index(begin = var_10427_begin_0, end = var_10427_end_0, end_mask = var_10427_end_mask_0, x = q_41_cast)[name = tensor<string, []>("op_10427_cast")];
            tensor<int32, [4]> var_10430_begin_0 = const()[name = tensor<string, []>("op_10430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10430_end_0 = const()[name = tensor<string, []>("op_10430_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10430_end_mask_0 = const()[name = tensor<string, []>("op_10430_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10430_cast = slice_by_index(begin = var_10430_begin_0, end = var_10430_end_0, end_mask = var_10430_end_mask_0, x = var_10391_cast)[name = tensor<string, []>("op_10430_cast")];
            tensor<int32, [4]> var_10431_begin_0 = const()[name = tensor<string, []>("op_10431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10431_end_0 = const()[name = tensor<string, []>("op_10431_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10431_end_mask_0 = const()[name = tensor<string, []>("op_10431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10431_cast = slice_by_index(begin = var_10431_begin_0, end = var_10431_end_0, end_mask = var_10431_end_mask_0, x = var_10391_cast)[name = tensor<string, []>("op_10431_cast")];
            tensor<int32, [4]> var_10432_begin_0 = const()[name = tensor<string, []>("op_10432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10432_end_0 = const()[name = tensor<string, []>("op_10432_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10432_end_mask_0 = const()[name = tensor<string, []>("op_10432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10432_cast = slice_by_index(begin = var_10432_begin_0, end = var_10432_end_0, end_mask = var_10432_end_mask_0, x = var_10395_cast)[name = tensor<string, []>("op_10432_cast")];
            tensor<int32, [4]> var_10433_begin_0 = const()[name = tensor<string, []>("op_10433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10433_end_0 = const()[name = tensor<string, []>("op_10433_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10433_end_mask_0 = const()[name = tensor<string, []>("op_10433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10433_cast = slice_by_index(begin = var_10433_begin_0, end = var_10433_end_0, end_mask = var_10433_end_mask_0, x = var_10395_cast)[name = tensor<string, []>("op_10433_cast")];
            tensor<int32, [4]> var_10434_begin_0 = const()[name = tensor<string, []>("op_10434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10434_end_0 = const()[name = tensor<string, []>("op_10434_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10434_end_mask_0 = const()[name = tensor<string, []>("op_10434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10434_cast = slice_by_index(begin = var_10434_begin_0, end = var_10434_end_0, end_mask = var_10434_end_mask_0, x = var_10399_cast)[name = tensor<string, []>("op_10434_cast")];
            tensor<int32, [4]> var_10435_begin_0 = const()[name = tensor<string, []>("op_10435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10435_end_0 = const()[name = tensor<string, []>("op_10435_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10435_end_mask_0 = const()[name = tensor<string, []>("op_10435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10435_cast = slice_by_index(begin = var_10435_begin_0, end = var_10435_end_0, end_mask = var_10435_end_mask_0, x = var_10399_cast)[name = tensor<string, []>("op_10435_cast")];
            tensor<int32, [4]> var_10436_begin_0 = const()[name = tensor<string, []>("op_10436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10436_end_0 = const()[name = tensor<string, []>("op_10436_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10436_end_mask_0 = const()[name = tensor<string, []>("op_10436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10436_cast = slice_by_index(begin = var_10436_begin_0, end = var_10436_end_0, end_mask = var_10436_end_mask_0, x = var_10403_cast)[name = tensor<string, []>("op_10436_cast")];
            tensor<int32, [4]> var_10437_begin_0 = const()[name = tensor<string, []>("op_10437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10437_end_0 = const()[name = tensor<string, []>("op_10437_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10437_end_mask_0 = const()[name = tensor<string, []>("op_10437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10437_cast = slice_by_index(begin = var_10437_begin_0, end = var_10437_end_0, end_mask = var_10437_end_mask_0, x = var_10403_cast)[name = tensor<string, []>("op_10437_cast")];
            tensor<int32, [4]> var_10438_begin_0 = const()[name = tensor<string, []>("op_10438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10438_end_0 = const()[name = tensor<string, []>("op_10438_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10438_end_mask_0 = const()[name = tensor<string, []>("op_10438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10438_cast = slice_by_index(begin = var_10438_begin_0, end = var_10438_end_0, end_mask = var_10438_end_mask_0, x = var_10407_cast)[name = tensor<string, []>("op_10438_cast")];
            tensor<int32, [4]> var_10439_begin_0 = const()[name = tensor<string, []>("op_10439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10439_end_0 = const()[name = tensor<string, []>("op_10439_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10439_end_mask_0 = const()[name = tensor<string, []>("op_10439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10439_cast = slice_by_index(begin = var_10439_begin_0, end = var_10439_end_0, end_mask = var_10439_end_mask_0, x = var_10407_cast)[name = tensor<string, []>("op_10439_cast")];
            tensor<int32, [4]> var_10440_begin_0 = const()[name = tensor<string, []>("op_10440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10440_end_0 = const()[name = tensor<string, []>("op_10440_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10440_end_mask_0 = const()[name = tensor<string, []>("op_10440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10440_cast = slice_by_index(begin = var_10440_begin_0, end = var_10440_end_0, end_mask = var_10440_end_mask_0, x = var_10411_cast)[name = tensor<string, []>("op_10440_cast")];
            tensor<int32, [4]> var_10441_begin_0 = const()[name = tensor<string, []>("op_10441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10441_end_0 = const()[name = tensor<string, []>("op_10441_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10441_end_mask_0 = const()[name = tensor<string, []>("op_10441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10441_cast = slice_by_index(begin = var_10441_begin_0, end = var_10441_end_0, end_mask = var_10441_end_mask_0, x = var_10411_cast)[name = tensor<string, []>("op_10441_cast")];
            tensor<int32, [4]> var_10442_begin_0 = const()[name = tensor<string, []>("op_10442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10442_end_0 = const()[name = tensor<string, []>("op_10442_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10442_end_mask_0 = const()[name = tensor<string, []>("op_10442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10442_cast = slice_by_index(begin = var_10442_begin_0, end = var_10442_end_0, end_mask = var_10442_end_mask_0, x = var_10415_cast)[name = tensor<string, []>("op_10442_cast")];
            tensor<int32, [4]> var_10443_begin_0 = const()[name = tensor<string, []>("op_10443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10443_end_0 = const()[name = tensor<string, []>("op_10443_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10443_end_mask_0 = const()[name = tensor<string, []>("op_10443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10443_cast = slice_by_index(begin = var_10443_begin_0, end = var_10443_end_0, end_mask = var_10443_end_mask_0, x = var_10415_cast)[name = tensor<string, []>("op_10443_cast")];
            tensor<int32, [4]> var_10444_begin_0 = const()[name = tensor<string, []>("op_10444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10444_end_0 = const()[name = tensor<string, []>("op_10444_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10444_end_mask_0 = const()[name = tensor<string, []>("op_10444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10444_cast = slice_by_index(begin = var_10444_begin_0, end = var_10444_end_0, end_mask = var_10444_end_mask_0, x = var_10419_cast)[name = tensor<string, []>("op_10444_cast")];
            tensor<int32, [4]> var_10445_begin_0 = const()[name = tensor<string, []>("op_10445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10445_end_0 = const()[name = tensor<string, []>("op_10445_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10445_end_mask_0 = const()[name = tensor<string, []>("op_10445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10445_cast = slice_by_index(begin = var_10445_begin_0, end = var_10445_end_0, end_mask = var_10445_end_mask_0, x = var_10419_cast)[name = tensor<string, []>("op_10445_cast")];
            tensor<int32, [4]> var_10446_begin_0 = const()[name = tensor<string, []>("op_10446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10446_end_0 = const()[name = tensor<string, []>("op_10446_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10446_end_mask_0 = const()[name = tensor<string, []>("op_10446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10446_cast = slice_by_index(begin = var_10446_begin_0, end = var_10446_end_0, end_mask = var_10446_end_mask_0, x = var_10423_cast)[name = tensor<string, []>("op_10446_cast")];
            tensor<int32, [4]> var_10447_begin_0 = const()[name = tensor<string, []>("op_10447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10447_end_0 = const()[name = tensor<string, []>("op_10447_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10447_end_mask_0 = const()[name = tensor<string, []>("op_10447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10447_cast = slice_by_index(begin = var_10447_begin_0, end = var_10447_end_0, end_mask = var_10447_end_mask_0, x = var_10423_cast)[name = tensor<string, []>("op_10447_cast")];
            tensor<int32, [4]> var_10448_begin_0 = const()[name = tensor<string, []>("op_10448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10448_end_0 = const()[name = tensor<string, []>("op_10448_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10448_end_mask_0 = const()[name = tensor<string, []>("op_10448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10448_cast = slice_by_index(begin = var_10448_begin_0, end = var_10448_end_0, end_mask = var_10448_end_mask_0, x = var_10427_cast)[name = tensor<string, []>("op_10448_cast")];
            tensor<int32, [4]> var_10449_begin_0 = const()[name = tensor<string, []>("op_10449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10449_end_0 = const()[name = tensor<string, []>("op_10449_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10449_end_mask_0 = const()[name = tensor<string, []>("op_10449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10449_cast = slice_by_index(begin = var_10449_begin_0, end = var_10449_end_0, end_mask = var_10449_end_mask_0, x = var_10427_cast)[name = tensor<string, []>("op_10449_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_10454_begin_0 = const()[name = tensor<string, []>("op_10454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10454_end_0 = const()[name = tensor<string, []>("op_10454_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_10454_end_mask_0 = const()[name = tensor<string, []>("op_10454_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, 64]> var_10454_cast = slice_by_index(begin = var_10454_begin_0, end = var_10454_end_0, end_mask = var_10454_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10454_cast")];
            tensor<int32, [4]> var_10458_begin_0 = const()[name = tensor<string, []>("op_10458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10458_end_0 = const()[name = tensor<string, []>("op_10458_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_10458_end_mask_0 = const()[name = tensor<string, []>("op_10458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10458_cast = slice_by_index(begin = var_10458_begin_0, end = var_10458_end_0, end_mask = var_10458_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10458_cast")];
            tensor<int32, [4]> var_10462_begin_0 = const()[name = tensor<string, []>("op_10462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10462_end_0 = const()[name = tensor<string, []>("op_10462_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_10462_end_mask_0 = const()[name = tensor<string, []>("op_10462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10462_cast = slice_by_index(begin = var_10462_begin_0, end = var_10462_end_0, end_mask = var_10462_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10462_cast")];
            tensor<int32, [4]> var_10466_begin_0 = const()[name = tensor<string, []>("op_10466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10466_end_0 = const()[name = tensor<string, []>("op_10466_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_10466_end_mask_0 = const()[name = tensor<string, []>("op_10466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10466_cast = slice_by_index(begin = var_10466_begin_0, end = var_10466_end_0, end_mask = var_10466_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10466_cast")];
            tensor<int32, [4]> var_10470_begin_0 = const()[name = tensor<string, []>("op_10470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10470_end_0 = const()[name = tensor<string, []>("op_10470_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_10470_end_mask_0 = const()[name = tensor<string, []>("op_10470_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10470_cast = slice_by_index(begin = var_10470_begin_0, end = var_10470_end_0, end_mask = var_10470_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10470_cast")];
            tensor<int32, [4]> var_10474_begin_0 = const()[name = tensor<string, []>("op_10474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10474_end_0 = const()[name = tensor<string, []>("op_10474_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_10474_end_mask_0 = const()[name = tensor<string, []>("op_10474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10474_cast = slice_by_index(begin = var_10474_begin_0, end = var_10474_end_0, end_mask = var_10474_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10474_cast")];
            tensor<int32, [4]> var_10478_begin_0 = const()[name = tensor<string, []>("op_10478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_10478_end_0 = const()[name = tensor<string, []>("op_10478_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_10478_end_mask_0 = const()[name = tensor<string, []>("op_10478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10478_cast = slice_by_index(begin = var_10478_begin_0, end = var_10478_end_0, end_mask = var_10478_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10478_cast")];
            tensor<int32, [4]> var_10482_begin_0 = const()[name = tensor<string, []>("op_10482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_10482_end_0 = const()[name = tensor<string, []>("op_10482_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_10482_end_mask_0 = const()[name = tensor<string, []>("op_10482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10482_cast = slice_by_index(begin = var_10482_begin_0, end = var_10482_end_0, end_mask = var_10482_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10482_cast")];
            tensor<int32, [4]> var_10486_begin_0 = const()[name = tensor<string, []>("op_10486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10486_end_0 = const()[name = tensor<string, []>("op_10486_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_10486_end_mask_0 = const()[name = tensor<string, []>("op_10486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10486_cast = slice_by_index(begin = var_10486_begin_0, end = var_10486_end_0, end_mask = var_10486_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10486_cast")];
            tensor<int32, [4]> var_10490_begin_0 = const()[name = tensor<string, []>("op_10490_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_10490_end_0 = const()[name = tensor<string, []>("op_10490_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_10490_end_mask_0 = const()[name = tensor<string, []>("op_10490_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_10490_cast = slice_by_index(begin = var_10490_begin_0, end = var_10490_end_0, end_mask = var_10490_end_mask_0, x = transpose_11)[name = tensor<string, []>("op_10490_cast")];
            tensor<int32, [4]> var_10492_begin_0 = const()[name = tensor<string, []>("op_10492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10492_end_0 = const()[name = tensor<string, []>("op_10492_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10492_end_mask_0 = const()[name = tensor<string, []>("op_10492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10492_cast = slice_by_index(begin = var_10492_begin_0, end = var_10492_end_0, end_mask = var_10492_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10492_cast")];
            tensor<int32, [4]> var_10496_begin_0 = const()[name = tensor<string, []>("op_10496_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10496_end_0 = const()[name = tensor<string, []>("op_10496_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10496_end_mask_0 = const()[name = tensor<string, []>("op_10496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10496_cast = slice_by_index(begin = var_10496_begin_0, end = var_10496_end_0, end_mask = var_10496_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10496_cast")];
            tensor<int32, [4]> var_10500_begin_0 = const()[name = tensor<string, []>("op_10500_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10500_end_0 = const()[name = tensor<string, []>("op_10500_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10500_end_mask_0 = const()[name = tensor<string, []>("op_10500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10500_cast = slice_by_index(begin = var_10500_begin_0, end = var_10500_end_0, end_mask = var_10500_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10500_cast")];
            tensor<int32, [4]> var_10504_begin_0 = const()[name = tensor<string, []>("op_10504_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10504_end_0 = const()[name = tensor<string, []>("op_10504_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10504_end_mask_0 = const()[name = tensor<string, []>("op_10504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10504_cast = slice_by_index(begin = var_10504_begin_0, end = var_10504_end_0, end_mask = var_10504_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10504_cast")];
            tensor<int32, [4]> var_10508_begin_0 = const()[name = tensor<string, []>("op_10508_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10508_end_0 = const()[name = tensor<string, []>("op_10508_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10508_end_mask_0 = const()[name = tensor<string, []>("op_10508_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10508_cast = slice_by_index(begin = var_10508_begin_0, end = var_10508_end_0, end_mask = var_10508_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10508_cast")];
            tensor<int32, [4]> var_10512_begin_0 = const()[name = tensor<string, []>("op_10512_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10512_end_0 = const()[name = tensor<string, []>("op_10512_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10512_end_mask_0 = const()[name = tensor<string, []>("op_10512_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10512_cast = slice_by_index(begin = var_10512_begin_0, end = var_10512_end_0, end_mask = var_10512_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10512_cast")];
            tensor<int32, [4]> var_10516_begin_0 = const()[name = tensor<string, []>("op_10516_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10516_end_0 = const()[name = tensor<string, []>("op_10516_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10516_end_mask_0 = const()[name = tensor<string, []>("op_10516_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10516_cast = slice_by_index(begin = var_10516_begin_0, end = var_10516_end_0, end_mask = var_10516_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10516_cast")];
            tensor<int32, [4]> var_10520_begin_0 = const()[name = tensor<string, []>("op_10520_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10520_end_0 = const()[name = tensor<string, []>("op_10520_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10520_end_mask_0 = const()[name = tensor<string, []>("op_10520_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10520_cast = slice_by_index(begin = var_10520_begin_0, end = var_10520_end_0, end_mask = var_10520_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10520_cast")];
            tensor<int32, [4]> var_10524_begin_0 = const()[name = tensor<string, []>("op_10524_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10524_end_0 = const()[name = tensor<string, []>("op_10524_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10524_end_mask_0 = const()[name = tensor<string, []>("op_10524_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10524_cast = slice_by_index(begin = var_10524_begin_0, end = var_10524_end_0, end_mask = var_10524_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10524_cast")];
            tensor<int32, [4]> var_10528_begin_0 = const()[name = tensor<string, []>("op_10528_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10528_end_0 = const()[name = tensor<string, []>("op_10528_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10528_end_mask_0 = const()[name = tensor<string, []>("op_10528_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10528_cast = slice_by_index(begin = var_10528_begin_0, end = var_10528_end_0, end_mask = var_10528_end_mask_0, x = v_41_cast)[name = tensor<string, []>("op_10528_cast")];
            tensor<string, []> var_10532_equation_0 = const()[name = tensor<string, []>("op_10532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10532_cast = einsum(equation = var_10532_equation_0, values = (var_10454_cast, var_10430_cast))[name = tensor<string, []>("op_10532_cast")];
            tensor<fp16, []> var_10533_to_fp16 = const()[name = tensor<string, []>("op_10533_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_481_cast = mul(x = var_10532_cast, y = var_10533_to_fp16)[name = tensor<string, []>("aw_chunk_481_cast")];
            tensor<string, []> var_10536_equation_0 = const()[name = tensor<string, []>("op_10536_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10536_cast = einsum(equation = var_10536_equation_0, values = (var_10454_cast, var_10431_cast))[name = tensor<string, []>("op_10536_cast")];
            tensor<fp16, []> var_10537_to_fp16 = const()[name = tensor<string, []>("op_10537_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_483_cast = mul(x = var_10536_cast, y = var_10537_to_fp16)[name = tensor<string, []>("aw_chunk_483_cast")];
            tensor<string, []> var_10540_equation_0 = const()[name = tensor<string, []>("op_10540_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10540_cast = einsum(equation = var_10540_equation_0, values = (var_10458_cast, var_10432_cast))[name = tensor<string, []>("op_10540_cast")];
            tensor<fp16, []> var_10541_to_fp16 = const()[name = tensor<string, []>("op_10541_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_485_cast = mul(x = var_10540_cast, y = var_10541_to_fp16)[name = tensor<string, []>("aw_chunk_485_cast")];
            tensor<string, []> var_10544_equation_0 = const()[name = tensor<string, []>("op_10544_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10544_cast = einsum(equation = var_10544_equation_0, values = (var_10458_cast, var_10433_cast))[name = tensor<string, []>("op_10544_cast")];
            tensor<fp16, []> var_10545_to_fp16 = const()[name = tensor<string, []>("op_10545_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_487_cast = mul(x = var_10544_cast, y = var_10545_to_fp16)[name = tensor<string, []>("aw_chunk_487_cast")];
            tensor<string, []> var_10548_equation_0 = const()[name = tensor<string, []>("op_10548_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10548_cast = einsum(equation = var_10548_equation_0, values = (var_10462_cast, var_10434_cast))[name = tensor<string, []>("op_10548_cast")];
            tensor<fp16, []> var_10549_to_fp16 = const()[name = tensor<string, []>("op_10549_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_489_cast = mul(x = var_10548_cast, y = var_10549_to_fp16)[name = tensor<string, []>("aw_chunk_489_cast")];
            tensor<string, []> var_10552_equation_0 = const()[name = tensor<string, []>("op_10552_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10552_cast = einsum(equation = var_10552_equation_0, values = (var_10462_cast, var_10435_cast))[name = tensor<string, []>("op_10552_cast")];
            tensor<fp16, []> var_10553_to_fp16 = const()[name = tensor<string, []>("op_10553_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_491_cast = mul(x = var_10552_cast, y = var_10553_to_fp16)[name = tensor<string, []>("aw_chunk_491_cast")];
            tensor<string, []> var_10556_equation_0 = const()[name = tensor<string, []>("op_10556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10556_cast = einsum(equation = var_10556_equation_0, values = (var_10466_cast, var_10436_cast))[name = tensor<string, []>("op_10556_cast")];
            tensor<fp16, []> var_10557_to_fp16 = const()[name = tensor<string, []>("op_10557_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_493_cast = mul(x = var_10556_cast, y = var_10557_to_fp16)[name = tensor<string, []>("aw_chunk_493_cast")];
            tensor<string, []> var_10560_equation_0 = const()[name = tensor<string, []>("op_10560_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10560_cast = einsum(equation = var_10560_equation_0, values = (var_10466_cast, var_10437_cast))[name = tensor<string, []>("op_10560_cast")];
            tensor<fp16, []> var_10561_to_fp16 = const()[name = tensor<string, []>("op_10561_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_495_cast = mul(x = var_10560_cast, y = var_10561_to_fp16)[name = tensor<string, []>("aw_chunk_495_cast")];
            tensor<string, []> var_10564_equation_0 = const()[name = tensor<string, []>("op_10564_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10564_cast = einsum(equation = var_10564_equation_0, values = (var_10470_cast, var_10438_cast))[name = tensor<string, []>("op_10564_cast")];
            tensor<fp16, []> var_10565_to_fp16 = const()[name = tensor<string, []>("op_10565_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_497_cast = mul(x = var_10564_cast, y = var_10565_to_fp16)[name = tensor<string, []>("aw_chunk_497_cast")];
            tensor<string, []> var_10568_equation_0 = const()[name = tensor<string, []>("op_10568_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10568_cast = einsum(equation = var_10568_equation_0, values = (var_10470_cast, var_10439_cast))[name = tensor<string, []>("op_10568_cast")];
            tensor<fp16, []> var_10569_to_fp16 = const()[name = tensor<string, []>("op_10569_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_499_cast = mul(x = var_10568_cast, y = var_10569_to_fp16)[name = tensor<string, []>("aw_chunk_499_cast")];
            tensor<string, []> var_10572_equation_0 = const()[name = tensor<string, []>("op_10572_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10572_cast = einsum(equation = var_10572_equation_0, values = (var_10474_cast, var_10440_cast))[name = tensor<string, []>("op_10572_cast")];
            tensor<fp16, []> var_10573_to_fp16 = const()[name = tensor<string, []>("op_10573_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_501_cast = mul(x = var_10572_cast, y = var_10573_to_fp16)[name = tensor<string, []>("aw_chunk_501_cast")];
            tensor<string, []> var_10576_equation_0 = const()[name = tensor<string, []>("op_10576_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10576_cast = einsum(equation = var_10576_equation_0, values = (var_10474_cast, var_10441_cast))[name = tensor<string, []>("op_10576_cast")];
            tensor<fp16, []> var_10577_to_fp16 = const()[name = tensor<string, []>("op_10577_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_503_cast = mul(x = var_10576_cast, y = var_10577_to_fp16)[name = tensor<string, []>("aw_chunk_503_cast")];
            tensor<string, []> var_10580_equation_0 = const()[name = tensor<string, []>("op_10580_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10580_cast = einsum(equation = var_10580_equation_0, values = (var_10478_cast, var_10442_cast))[name = tensor<string, []>("op_10580_cast")];
            tensor<fp16, []> var_10581_to_fp16 = const()[name = tensor<string, []>("op_10581_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_505_cast = mul(x = var_10580_cast, y = var_10581_to_fp16)[name = tensor<string, []>("aw_chunk_505_cast")];
            tensor<string, []> var_10584_equation_0 = const()[name = tensor<string, []>("op_10584_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10584_cast = einsum(equation = var_10584_equation_0, values = (var_10478_cast, var_10443_cast))[name = tensor<string, []>("op_10584_cast")];
            tensor<fp16, []> var_10585_to_fp16 = const()[name = tensor<string, []>("op_10585_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_507_cast = mul(x = var_10584_cast, y = var_10585_to_fp16)[name = tensor<string, []>("aw_chunk_507_cast")];
            tensor<string, []> var_10588_equation_0 = const()[name = tensor<string, []>("op_10588_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10588_cast = einsum(equation = var_10588_equation_0, values = (var_10482_cast, var_10444_cast))[name = tensor<string, []>("op_10588_cast")];
            tensor<fp16, []> var_10589_to_fp16 = const()[name = tensor<string, []>("op_10589_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_509_cast = mul(x = var_10588_cast, y = var_10589_to_fp16)[name = tensor<string, []>("aw_chunk_509_cast")];
            tensor<string, []> var_10592_equation_0 = const()[name = tensor<string, []>("op_10592_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10592_cast = einsum(equation = var_10592_equation_0, values = (var_10482_cast, var_10445_cast))[name = tensor<string, []>("op_10592_cast")];
            tensor<fp16, []> var_10593_to_fp16 = const()[name = tensor<string, []>("op_10593_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_511_cast = mul(x = var_10592_cast, y = var_10593_to_fp16)[name = tensor<string, []>("aw_chunk_511_cast")];
            tensor<string, []> var_10596_equation_0 = const()[name = tensor<string, []>("op_10596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10596_cast = einsum(equation = var_10596_equation_0, values = (var_10486_cast, var_10446_cast))[name = tensor<string, []>("op_10596_cast")];
            tensor<fp16, []> var_10597_to_fp16 = const()[name = tensor<string, []>("op_10597_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_513_cast = mul(x = var_10596_cast, y = var_10597_to_fp16)[name = tensor<string, []>("aw_chunk_513_cast")];
            tensor<string, []> var_10600_equation_0 = const()[name = tensor<string, []>("op_10600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10600_cast = einsum(equation = var_10600_equation_0, values = (var_10486_cast, var_10447_cast))[name = tensor<string, []>("op_10600_cast")];
            tensor<fp16, []> var_10601_to_fp16 = const()[name = tensor<string, []>("op_10601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_515_cast = mul(x = var_10600_cast, y = var_10601_to_fp16)[name = tensor<string, []>("aw_chunk_515_cast")];
            tensor<string, []> var_10604_equation_0 = const()[name = tensor<string, []>("op_10604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10604_cast = einsum(equation = var_10604_equation_0, values = (var_10490_cast, var_10448_cast))[name = tensor<string, []>("op_10604_cast")];
            tensor<fp16, []> var_10605_to_fp16 = const()[name = tensor<string, []>("op_10605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_517_cast = mul(x = var_10604_cast, y = var_10605_to_fp16)[name = tensor<string, []>("aw_chunk_517_cast")];
            tensor<string, []> var_10608_equation_0 = const()[name = tensor<string, []>("op_10608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_10608_cast = einsum(equation = var_10608_equation_0, values = (var_10490_cast, var_10449_cast))[name = tensor<string, []>("op_10608_cast")];
            tensor<fp16, []> var_10609_to_fp16 = const()[name = tensor<string, []>("op_10609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_519_cast = mul(x = var_10608_cast, y = var_10609_to_fp16)[name = tensor<string, []>("aw_chunk_519_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10611_cast = softmax(axis = var_10257, x = aw_chunk_481_cast)[name = tensor<string, []>("op_10611_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10612_cast = softmax(axis = var_10257, x = aw_chunk_483_cast)[name = tensor<string, []>("op_10612_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10613_cast = softmax(axis = var_10257, x = aw_chunk_485_cast)[name = tensor<string, []>("op_10613_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10614_cast = softmax(axis = var_10257, x = aw_chunk_487_cast)[name = tensor<string, []>("op_10614_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10615_cast = softmax(axis = var_10257, x = aw_chunk_489_cast)[name = tensor<string, []>("op_10615_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10616_cast = softmax(axis = var_10257, x = aw_chunk_491_cast)[name = tensor<string, []>("op_10616_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10617_cast = softmax(axis = var_10257, x = aw_chunk_493_cast)[name = tensor<string, []>("op_10617_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10618_cast = softmax(axis = var_10257, x = aw_chunk_495_cast)[name = tensor<string, []>("op_10618_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10619_cast = softmax(axis = var_10257, x = aw_chunk_497_cast)[name = tensor<string, []>("op_10619_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10620_cast = softmax(axis = var_10257, x = aw_chunk_499_cast)[name = tensor<string, []>("op_10620_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10621_cast = softmax(axis = var_10257, x = aw_chunk_501_cast)[name = tensor<string, []>("op_10621_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10622_cast = softmax(axis = var_10257, x = aw_chunk_503_cast)[name = tensor<string, []>("op_10622_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10623_cast = softmax(axis = var_10257, x = aw_chunk_505_cast)[name = tensor<string, []>("op_10623_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10624_cast = softmax(axis = var_10257, x = aw_chunk_507_cast)[name = tensor<string, []>("op_10624_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10625_cast = softmax(axis = var_10257, x = aw_chunk_509_cast)[name = tensor<string, []>("op_10625_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10626_cast = softmax(axis = var_10257, x = aw_chunk_511_cast)[name = tensor<string, []>("op_10626_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10627_cast = softmax(axis = var_10257, x = aw_chunk_513_cast)[name = tensor<string, []>("op_10627_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10628_cast = softmax(axis = var_10257, x = aw_chunk_515_cast)[name = tensor<string, []>("op_10628_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10629_cast = softmax(axis = var_10257, x = aw_chunk_517_cast)[name = tensor<string, []>("op_10629_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_10630_cast = softmax(axis = var_10257, x = aw_chunk_519_cast)[name = tensor<string, []>("op_10630_cast")];
            tensor<string, []> var_10632_equation_0 = const()[name = tensor<string, []>("op_10632_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10632_cast = einsum(equation = var_10632_equation_0, values = (var_10492_cast, var_10611_cast))[name = tensor<string, []>("op_10632_cast")];
            tensor<string, []> var_10634_equation_0 = const()[name = tensor<string, []>("op_10634_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10634_cast = einsum(equation = var_10634_equation_0, values = (var_10492_cast, var_10612_cast))[name = tensor<string, []>("op_10634_cast")];
            tensor<string, []> var_10636_equation_0 = const()[name = tensor<string, []>("op_10636_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10636_cast = einsum(equation = var_10636_equation_0, values = (var_10496_cast, var_10613_cast))[name = tensor<string, []>("op_10636_cast")];
            tensor<string, []> var_10638_equation_0 = const()[name = tensor<string, []>("op_10638_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10638_cast = einsum(equation = var_10638_equation_0, values = (var_10496_cast, var_10614_cast))[name = tensor<string, []>("op_10638_cast")];
            tensor<string, []> var_10640_equation_0 = const()[name = tensor<string, []>("op_10640_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10640_cast = einsum(equation = var_10640_equation_0, values = (var_10500_cast, var_10615_cast))[name = tensor<string, []>("op_10640_cast")];
            tensor<string, []> var_10642_equation_0 = const()[name = tensor<string, []>("op_10642_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10642_cast = einsum(equation = var_10642_equation_0, values = (var_10500_cast, var_10616_cast))[name = tensor<string, []>("op_10642_cast")];
            tensor<string, []> var_10644_equation_0 = const()[name = tensor<string, []>("op_10644_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10644_cast = einsum(equation = var_10644_equation_0, values = (var_10504_cast, var_10617_cast))[name = tensor<string, []>("op_10644_cast")];
            tensor<string, []> var_10646_equation_0 = const()[name = tensor<string, []>("op_10646_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10646_cast = einsum(equation = var_10646_equation_0, values = (var_10504_cast, var_10618_cast))[name = tensor<string, []>("op_10646_cast")];
            tensor<string, []> var_10648_equation_0 = const()[name = tensor<string, []>("op_10648_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10648_cast = einsum(equation = var_10648_equation_0, values = (var_10508_cast, var_10619_cast))[name = tensor<string, []>("op_10648_cast")];
            tensor<string, []> var_10650_equation_0 = const()[name = tensor<string, []>("op_10650_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10650_cast = einsum(equation = var_10650_equation_0, values = (var_10508_cast, var_10620_cast))[name = tensor<string, []>("op_10650_cast")];
            tensor<string, []> var_10652_equation_0 = const()[name = tensor<string, []>("op_10652_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10652_cast = einsum(equation = var_10652_equation_0, values = (var_10512_cast, var_10621_cast))[name = tensor<string, []>("op_10652_cast")];
            tensor<string, []> var_10654_equation_0 = const()[name = tensor<string, []>("op_10654_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10654_cast = einsum(equation = var_10654_equation_0, values = (var_10512_cast, var_10622_cast))[name = tensor<string, []>("op_10654_cast")];
            tensor<string, []> var_10656_equation_0 = const()[name = tensor<string, []>("op_10656_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10656_cast = einsum(equation = var_10656_equation_0, values = (var_10516_cast, var_10623_cast))[name = tensor<string, []>("op_10656_cast")];
            tensor<string, []> var_10658_equation_0 = const()[name = tensor<string, []>("op_10658_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10658_cast = einsum(equation = var_10658_equation_0, values = (var_10516_cast, var_10624_cast))[name = tensor<string, []>("op_10658_cast")];
            tensor<string, []> var_10660_equation_0 = const()[name = tensor<string, []>("op_10660_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10660_cast = einsum(equation = var_10660_equation_0, values = (var_10520_cast, var_10625_cast))[name = tensor<string, []>("op_10660_cast")];
            tensor<string, []> var_10662_equation_0 = const()[name = tensor<string, []>("op_10662_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10662_cast = einsum(equation = var_10662_equation_0, values = (var_10520_cast, var_10626_cast))[name = tensor<string, []>("op_10662_cast")];
            tensor<string, []> var_10664_equation_0 = const()[name = tensor<string, []>("op_10664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10664_cast = einsum(equation = var_10664_equation_0, values = (var_10524_cast, var_10627_cast))[name = tensor<string, []>("op_10664_cast")];
            tensor<string, []> var_10666_equation_0 = const()[name = tensor<string, []>("op_10666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10666_cast = einsum(equation = var_10666_equation_0, values = (var_10524_cast, var_10628_cast))[name = tensor<string, []>("op_10666_cast")];
            tensor<string, []> var_10668_equation_0 = const()[name = tensor<string, []>("op_10668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10668_cast = einsum(equation = var_10668_equation_0, values = (var_10528_cast, var_10629_cast))[name = tensor<string, []>("op_10668_cast")];
            tensor<string, []> var_10670_equation_0 = const()[name = tensor<string, []>("op_10670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10670_cast = einsum(equation = var_10670_equation_0, values = (var_10528_cast, var_10630_cast))[name = tensor<string, []>("op_10670_cast")];
            tensor<bool, []> var_10672_interleave_0 = const()[name = tensor<string, []>("op_10672_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10672_cast = concat(axis = var_10234, interleave = var_10672_interleave_0, values = (var_10632_cast, var_10634_cast))[name = tensor<string, []>("op_10672_cast")];
            tensor<bool, []> var_10674_interleave_0 = const()[name = tensor<string, []>("op_10674_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10674_cast = concat(axis = var_10234, interleave = var_10674_interleave_0, values = (var_10636_cast, var_10638_cast))[name = tensor<string, []>("op_10674_cast")];
            tensor<bool, []> var_10676_interleave_0 = const()[name = tensor<string, []>("op_10676_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10676_cast = concat(axis = var_10234, interleave = var_10676_interleave_0, values = (var_10640_cast, var_10642_cast))[name = tensor<string, []>("op_10676_cast")];
            tensor<bool, []> var_10678_interleave_0 = const()[name = tensor<string, []>("op_10678_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10678_cast = concat(axis = var_10234, interleave = var_10678_interleave_0, values = (var_10644_cast, var_10646_cast))[name = tensor<string, []>("op_10678_cast")];
            tensor<bool, []> var_10680_interleave_0 = const()[name = tensor<string, []>("op_10680_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10680_cast = concat(axis = var_10234, interleave = var_10680_interleave_0, values = (var_10648_cast, var_10650_cast))[name = tensor<string, []>("op_10680_cast")];
            tensor<bool, []> var_10682_interleave_0 = const()[name = tensor<string, []>("op_10682_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10682_cast = concat(axis = var_10234, interleave = var_10682_interleave_0, values = (var_10652_cast, var_10654_cast))[name = tensor<string, []>("op_10682_cast")];
            tensor<bool, []> var_10684_interleave_0 = const()[name = tensor<string, []>("op_10684_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10684_cast = concat(axis = var_10234, interleave = var_10684_interleave_0, values = (var_10656_cast, var_10658_cast))[name = tensor<string, []>("op_10684_cast")];
            tensor<bool, []> var_10686_interleave_0 = const()[name = tensor<string, []>("op_10686_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10686_cast = concat(axis = var_10234, interleave = var_10686_interleave_0, values = (var_10660_cast, var_10662_cast))[name = tensor<string, []>("op_10686_cast")];
            tensor<bool, []> var_10688_interleave_0 = const()[name = tensor<string, []>("op_10688_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10688_cast = concat(axis = var_10234, interleave = var_10688_interleave_0, values = (var_10664_cast, var_10666_cast))[name = tensor<string, []>("op_10688_cast")];
            tensor<bool, []> var_10690_interleave_0 = const()[name = tensor<string, []>("op_10690_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_10690_cast = concat(axis = var_10234, interleave = var_10690_interleave_0, values = (var_10668_cast, var_10670_cast))[name = tensor<string, []>("op_10690_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_10257, interleave = input_383_interleave_0, values = (var_10672_cast, var_10674_cast, var_10676_cast, var_10678_cast, var_10680_cast, var_10682_cast, var_10684_cast, var_10686_cast, var_10688_cast, var_10690_cast))[name = tensor<string, []>("input_383_cast")];
            tensor<int32, [2]> var_10696 = const()[name = tensor<string, []>("op_10696"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10698 = const()[name = tensor<string, []>("op_10698"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_10700_pad_type_0 = const()[name = tensor<string, []>("op_10700_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_10700_pad_0 = const()[name = tensor<string, []>("op_10700_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198665920))), lut = tensor<fp16, [4]>([-0x1.9d4p-4, -0x1.ec8p-6, 0x1.eap-6, 0x1.9a4p-4]), 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, []>(198768384)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10700_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_10698, groups = var_10257, pad = var_10700_pad_0, pad_type = var_10700_pad_type_0, strides = var_10696, 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_10700_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_63_cast = add(x = var_10700_cast, y = inputs_61_cast)[name = tensor<string, []>("inputs_63_cast")];
            tensor<int32, [1]> var_10704 = const()[name = tensor<string, []>("op_10704"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_63_cast = reduce_mean(axes = var_10704, keep_dims = var_10252, 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_10708 = const()[name = tensor<string, []>("op_10708"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_10709_cast = reduce_mean(axes = var_10708, keep_dims = var_10252, x = zero_mean_sq_63_cast)[name = tensor<string, []>("op_10709_cast")];
            tensor<fp16, []> var_10710_to_fp16 = const()[name = tensor<string, []>("op_10710_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_10711_cast = add(x = var_10709_cast, y = var_10710_to_fp16)[name = tensor<string, []>("op_10711_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_10711_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_10715_to_fp16 = const()[name = tensor<string, []>("op_10715_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198769728)))];
            tensor<fp16, [2, 640, 1, 1024]> var_10716_cast = add(x = out_63_cast, y = var_10715_to_fp16)[name = tensor<string, []>("op_10716_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_10718_to_fp16 = const()[name = tensor<string, []>("op_10718_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198771072)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_229_cast = mul(x = var_10716_cast, y = var_10718_to_fp16)[name = tensor<string, []>("hidden_states_229_cast")];
            tensor<int32, [2]> var_10725 = const()[name = tensor<string, []>("op_10725"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10727 = const()[name = tensor<string, []>("op_10727"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198772416))), lut = tensor<fp16, [4]>([-0x1.888p-4, -0x1.d58p-6, 0x1.d68p-6, 0x1.888p-4]), 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_10727, groups = var_10257, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = var_10725, 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_10731 = const()[name = tensor<string, []>("op_10731"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10733 = const()[name = tensor<string, []>("op_10733"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_85_pad_type_0 = const()[name = tensor<string, []>("k_85_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_85_pad_0 = const()[name = tensor<string, []>("k_85_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(198874880))), lut = tensor<fp16, [4]>([-0x1.67p-4, -0x1.ad8p-6, 0x1.a1cp-6, 0x1.63cp-4]), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_85_cast = conv(dilations = var_10733, groups = var_10257, pad = k_85_pad_0, pad_type = k_85_pad_type_0, strides = var_10731, 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_10737 = const()[name = tensor<string, []>("op_10737"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_10739 = const()[name = tensor<string, []>("op_10739"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_43_pad_type_0 = const()[name = tensor<string, []>("v_43_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_43_pad_0 = const()[name = tensor<string, []>("v_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199038784))), lut = tensor<fp16, [4]>([-0x1.41p-4, -0x1.71cp-6, 0x1.6dp-6, 0x1.3fcp-4]), name = tensor<string, []>("up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_43_cast = conv(dilations = var_10739, groups = var_10257, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = var_10737, 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_10743_begin_0 = const()[name = tensor<string, []>("op_10743_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10743_end_0 = const()[name = tensor<string, []>("op_10743_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10743_end_mask_0 = const()[name = tensor<string, []>("op_10743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10743_cast = slice_by_index(begin = var_10743_begin_0, end = var_10743_end_0, end_mask = var_10743_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10743_cast")];
            tensor<int32, [4]> var_10747_begin_0 = const()[name = tensor<string, []>("op_10747_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10747_end_0 = const()[name = tensor<string, []>("op_10747_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_10747_end_mask_0 = const()[name = tensor<string, []>("op_10747_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10747_cast = slice_by_index(begin = var_10747_begin_0, end = var_10747_end_0, end_mask = var_10747_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10747_cast")];
            tensor<int32, [4]> var_10751_begin_0 = const()[name = tensor<string, []>("op_10751_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10751_end_0 = const()[name = tensor<string, []>("op_10751_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_10751_end_mask_0 = const()[name = tensor<string, []>("op_10751_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10751_cast = slice_by_index(begin = var_10751_begin_0, end = var_10751_end_0, end_mask = var_10751_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10751_cast")];
            tensor<int32, [4]> var_10755_begin_0 = const()[name = tensor<string, []>("op_10755_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10755_end_0 = const()[name = tensor<string, []>("op_10755_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_10755_end_mask_0 = const()[name = tensor<string, []>("op_10755_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10755_cast = slice_by_index(begin = var_10755_begin_0, end = var_10755_end_0, end_mask = var_10755_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10755_cast")];
            tensor<int32, [4]> var_10759_begin_0 = const()[name = tensor<string, []>("op_10759_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10759_end_0 = const()[name = tensor<string, []>("op_10759_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_10759_end_mask_0 = const()[name = tensor<string, []>("op_10759_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10759_cast = slice_by_index(begin = var_10759_begin_0, end = var_10759_end_0, end_mask = var_10759_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10759_cast")];
            tensor<int32, [4]> var_10763_begin_0 = const()[name = tensor<string, []>("op_10763_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10763_end_0 = const()[name = tensor<string, []>("op_10763_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_10763_end_mask_0 = const()[name = tensor<string, []>("op_10763_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10763_cast = slice_by_index(begin = var_10763_begin_0, end = var_10763_end_0, end_mask = var_10763_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10763_cast")];
            tensor<int32, [4]> var_10767_begin_0 = const()[name = tensor<string, []>("op_10767_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10767_end_0 = const()[name = tensor<string, []>("op_10767_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_10767_end_mask_0 = const()[name = tensor<string, []>("op_10767_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10767_cast = slice_by_index(begin = var_10767_begin_0, end = var_10767_end_0, end_mask = var_10767_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10767_cast")];
            tensor<int32, [4]> var_10771_begin_0 = const()[name = tensor<string, []>("op_10771_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10771_end_0 = const()[name = tensor<string, []>("op_10771_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_10771_end_mask_0 = const()[name = tensor<string, []>("op_10771_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10771_cast = slice_by_index(begin = var_10771_begin_0, end = var_10771_end_0, end_mask = var_10771_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10771_cast")];
            tensor<int32, [4]> var_10775_begin_0 = const()[name = tensor<string, []>("op_10775_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10775_end_0 = const()[name = tensor<string, []>("op_10775_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_10775_end_mask_0 = const()[name = tensor<string, []>("op_10775_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10775_cast = slice_by_index(begin = var_10775_begin_0, end = var_10775_end_0, end_mask = var_10775_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10775_cast")];
            tensor<int32, [4]> var_10779_begin_0 = const()[name = tensor<string, []>("op_10779_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10779_end_0 = const()[name = tensor<string, []>("op_10779_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_10779_end_mask_0 = const()[name = tensor<string, []>("op_10779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_10779_cast = slice_by_index(begin = var_10779_begin_0, end = var_10779_end_0, end_mask = var_10779_end_mask_0, x = q_43_cast)[name = tensor<string, []>("op_10779_cast")];
            tensor<int32, [4]> var_10782_begin_0 = const()[name = tensor<string, []>("op_10782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10782_end_0 = const()[name = tensor<string, []>("op_10782_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10782_end_mask_0 = const()[name = tensor<string, []>("op_10782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10782_cast = slice_by_index(begin = var_10782_begin_0, end = var_10782_end_0, end_mask = var_10782_end_mask_0, x = var_10743_cast)[name = tensor<string, []>("op_10782_cast")];
            tensor<int32, [4]> var_10783_begin_0 = const()[name = tensor<string, []>("op_10783_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10783_end_0 = const()[name = tensor<string, []>("op_10783_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10783_end_mask_0 = const()[name = tensor<string, []>("op_10783_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10783_cast = slice_by_index(begin = var_10783_begin_0, end = var_10783_end_0, end_mask = var_10783_end_mask_0, x = var_10743_cast)[name = tensor<string, []>("op_10783_cast")];
            tensor<int32, [4]> var_10784_begin_0 = const()[name = tensor<string, []>("op_10784_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10784_end_0 = const()[name = tensor<string, []>("op_10784_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10784_end_mask_0 = const()[name = tensor<string, []>("op_10784_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10784_cast = slice_by_index(begin = var_10784_begin_0, end = var_10784_end_0, end_mask = var_10784_end_mask_0, x = var_10747_cast)[name = tensor<string, []>("op_10784_cast")];
            tensor<int32, [4]> var_10785_begin_0 = const()[name = tensor<string, []>("op_10785_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10785_end_0 = const()[name = tensor<string, []>("op_10785_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10785_end_mask_0 = const()[name = tensor<string, []>("op_10785_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10785_cast = slice_by_index(begin = var_10785_begin_0, end = var_10785_end_0, end_mask = var_10785_end_mask_0, x = var_10747_cast)[name = tensor<string, []>("op_10785_cast")];
            tensor<int32, [4]> var_10786_begin_0 = const()[name = tensor<string, []>("op_10786_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10786_end_0 = const()[name = tensor<string, []>("op_10786_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10786_end_mask_0 = const()[name = tensor<string, []>("op_10786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10786_cast = slice_by_index(begin = var_10786_begin_0, end = var_10786_end_0, end_mask = var_10786_end_mask_0, x = var_10751_cast)[name = tensor<string, []>("op_10786_cast")];
            tensor<int32, [4]> var_10787_begin_0 = const()[name = tensor<string, []>("op_10787_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10787_end_0 = const()[name = tensor<string, []>("op_10787_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10787_end_mask_0 = const()[name = tensor<string, []>("op_10787_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10787_cast = slice_by_index(begin = var_10787_begin_0, end = var_10787_end_0, end_mask = var_10787_end_mask_0, x = var_10751_cast)[name = tensor<string, []>("op_10787_cast")];
            tensor<int32, [4]> var_10788_begin_0 = const()[name = tensor<string, []>("op_10788_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10788_end_0 = const()[name = tensor<string, []>("op_10788_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10788_end_mask_0 = const()[name = tensor<string, []>("op_10788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10788_cast = slice_by_index(begin = var_10788_begin_0, end = var_10788_end_0, end_mask = var_10788_end_mask_0, x = var_10755_cast)[name = tensor<string, []>("op_10788_cast")];
            tensor<int32, [4]> var_10789_begin_0 = const()[name = tensor<string, []>("op_10789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10789_end_0 = const()[name = tensor<string, []>("op_10789_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10789_end_mask_0 = const()[name = tensor<string, []>("op_10789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10789_cast = slice_by_index(begin = var_10789_begin_0, end = var_10789_end_0, end_mask = var_10789_end_mask_0, x = var_10755_cast)[name = tensor<string, []>("op_10789_cast")];
            tensor<int32, [4]> var_10790_begin_0 = const()[name = tensor<string, []>("op_10790_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10790_end_0 = const()[name = tensor<string, []>("op_10790_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10790_end_mask_0 = const()[name = tensor<string, []>("op_10790_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10790_cast = slice_by_index(begin = var_10790_begin_0, end = var_10790_end_0, end_mask = var_10790_end_mask_0, x = var_10759_cast)[name = tensor<string, []>("op_10790_cast")];
            tensor<int32, [4]> var_10791_begin_0 = const()[name = tensor<string, []>("op_10791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10791_end_0 = const()[name = tensor<string, []>("op_10791_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10791_end_mask_0 = const()[name = tensor<string, []>("op_10791_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10791_cast = slice_by_index(begin = var_10791_begin_0, end = var_10791_end_0, end_mask = var_10791_end_mask_0, x = var_10759_cast)[name = tensor<string, []>("op_10791_cast")];
            tensor<int32, [4]> var_10792_begin_0 = const()[name = tensor<string, []>("op_10792_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10792_end_0 = const()[name = tensor<string, []>("op_10792_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10792_end_mask_0 = const()[name = tensor<string, []>("op_10792_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10792_cast = slice_by_index(begin = var_10792_begin_0, end = var_10792_end_0, end_mask = var_10792_end_mask_0, x = var_10763_cast)[name = tensor<string, []>("op_10792_cast")];
            tensor<int32, [4]> var_10793_begin_0 = const()[name = tensor<string, []>("op_10793_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10793_end_0 = const()[name = tensor<string, []>("op_10793_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10793_end_mask_0 = const()[name = tensor<string, []>("op_10793_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10793_cast = slice_by_index(begin = var_10793_begin_0, end = var_10793_end_0, end_mask = var_10793_end_mask_0, x = var_10763_cast)[name = tensor<string, []>("op_10793_cast")];
            tensor<int32, [4]> var_10794_begin_0 = const()[name = tensor<string, []>("op_10794_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10794_end_0 = const()[name = tensor<string, []>("op_10794_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10794_end_mask_0 = const()[name = tensor<string, []>("op_10794_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10794_cast = slice_by_index(begin = var_10794_begin_0, end = var_10794_end_0, end_mask = var_10794_end_mask_0, x = var_10767_cast)[name = tensor<string, []>("op_10794_cast")];
            tensor<int32, [4]> var_10795_begin_0 = const()[name = tensor<string, []>("op_10795_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10795_end_0 = const()[name = tensor<string, []>("op_10795_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10795_end_mask_0 = const()[name = tensor<string, []>("op_10795_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10795_cast = slice_by_index(begin = var_10795_begin_0, end = var_10795_end_0, end_mask = var_10795_end_mask_0, x = var_10767_cast)[name = tensor<string, []>("op_10795_cast")];
            tensor<int32, [4]> var_10796_begin_0 = const()[name = tensor<string, []>("op_10796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10796_end_0 = const()[name = tensor<string, []>("op_10796_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10796_end_mask_0 = const()[name = tensor<string, []>("op_10796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10796_cast = slice_by_index(begin = var_10796_begin_0, end = var_10796_end_0, end_mask = var_10796_end_mask_0, x = var_10771_cast)[name = tensor<string, []>("op_10796_cast")];
            tensor<int32, [4]> var_10797_begin_0 = const()[name = tensor<string, []>("op_10797_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10797_end_0 = const()[name = tensor<string, []>("op_10797_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10797_end_mask_0 = const()[name = tensor<string, []>("op_10797_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10797_cast = slice_by_index(begin = var_10797_begin_0, end = var_10797_end_0, end_mask = var_10797_end_mask_0, x = var_10771_cast)[name = tensor<string, []>("op_10797_cast")];
            tensor<int32, [4]> var_10798_begin_0 = const()[name = tensor<string, []>("op_10798_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10798_end_0 = const()[name = tensor<string, []>("op_10798_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10798_end_mask_0 = const()[name = tensor<string, []>("op_10798_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10798_cast = slice_by_index(begin = var_10798_begin_0, end = var_10798_end_0, end_mask = var_10798_end_mask_0, x = var_10775_cast)[name = tensor<string, []>("op_10798_cast")];
            tensor<int32, [4]> var_10799_begin_0 = const()[name = tensor<string, []>("op_10799_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10799_end_0 = const()[name = tensor<string, []>("op_10799_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10799_end_mask_0 = const()[name = tensor<string, []>("op_10799_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10799_cast = slice_by_index(begin = var_10799_begin_0, end = var_10799_end_0, end_mask = var_10799_end_mask_0, x = var_10775_cast)[name = tensor<string, []>("op_10799_cast")];
            tensor<int32, [4]> var_10800_begin_0 = const()[name = tensor<string, []>("op_10800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10800_end_0 = const()[name = tensor<string, []>("op_10800_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_10800_end_mask_0 = const()[name = tensor<string, []>("op_10800_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10800_cast = slice_by_index(begin = var_10800_begin_0, end = var_10800_end_0, end_mask = var_10800_end_mask_0, x = var_10779_cast)[name = tensor<string, []>("op_10800_cast")];
            tensor<int32, [4]> var_10801_begin_0 = const()[name = tensor<string, []>("op_10801_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10801_end_0 = const()[name = tensor<string, []>("op_10801_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_10801_end_mask_0 = const()[name = tensor<string, []>("op_10801_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_10801_cast = slice_by_index(begin = var_10801_begin_0, end = var_10801_end_0, end_mask = var_10801_end_mask_0, x = var_10779_cast)[name = tensor<string, []>("op_10801_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_10806_begin_0 = const()[name = tensor<string, []>("op_10806_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10806_end_0 = const()[name = tensor<string, []>("op_10806_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_10806_end_mask_0 = const()[name = tensor<string, []>("op_10806_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, 64]> var_10806_cast = slice_by_index(begin = var_10806_begin_0, end = var_10806_end_0, end_mask = var_10806_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10806_cast")];
            tensor<int32, [4]> var_10810_begin_0 = const()[name = tensor<string, []>("op_10810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_10810_end_0 = const()[name = tensor<string, []>("op_10810_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_10810_end_mask_0 = const()[name = tensor<string, []>("op_10810_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10810_cast = slice_by_index(begin = var_10810_begin_0, end = var_10810_end_0, end_mask = var_10810_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10810_cast")];
            tensor<int32, [4]> var_10814_begin_0 = const()[name = tensor<string, []>("op_10814_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_10814_end_0 = const()[name = tensor<string, []>("op_10814_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_10814_end_mask_0 = const()[name = tensor<string, []>("op_10814_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10814_cast = slice_by_index(begin = var_10814_begin_0, end = var_10814_end_0, end_mask = var_10814_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10814_cast")];
            tensor<int32, [4]> var_10818_begin_0 = const()[name = tensor<string, []>("op_10818_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_10818_end_0 = const()[name = tensor<string, []>("op_10818_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_10818_end_mask_0 = const()[name = tensor<string, []>("op_10818_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10818_cast = slice_by_index(begin = var_10818_begin_0, end = var_10818_end_0, end_mask = var_10818_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10818_cast")];
            tensor<int32, [4]> var_10822_begin_0 = const()[name = tensor<string, []>("op_10822_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_10822_end_0 = const()[name = tensor<string, []>("op_10822_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_10822_end_mask_0 = const()[name = tensor<string, []>("op_10822_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10822_cast = slice_by_index(begin = var_10822_begin_0, end = var_10822_end_0, end_mask = var_10822_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10822_cast")];
            tensor<int32, [4]> var_10826_begin_0 = const()[name = tensor<string, []>("op_10826_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_10826_end_0 = const()[name = tensor<string, []>("op_10826_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_10826_end_mask_0 = const()[name = tensor<string, []>("op_10826_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10826_cast = slice_by_index(begin = var_10826_begin_0, end = var_10826_end_0, end_mask = var_10826_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10826_cast")];
            tensor<int32, [4]> var_10830_begin_0 = const()[name = tensor<string, []>("op_10830_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_10830_end_0 = const()[name = tensor<string, []>("op_10830_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_10830_end_mask_0 = const()[name = tensor<string, []>("op_10830_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10830_cast = slice_by_index(begin = var_10830_begin_0, end = var_10830_end_0, end_mask = var_10830_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10830_cast")];
            tensor<int32, [4]> var_10834_begin_0 = const()[name = tensor<string, []>("op_10834_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_10834_end_0 = const()[name = tensor<string, []>("op_10834_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_10834_end_mask_0 = const()[name = tensor<string, []>("op_10834_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10834_cast = slice_by_index(begin = var_10834_begin_0, end = var_10834_end_0, end_mask = var_10834_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10834_cast")];
            tensor<int32, [4]> var_10838_begin_0 = const()[name = tensor<string, []>("op_10838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_10838_end_0 = const()[name = tensor<string, []>("op_10838_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_10838_end_mask_0 = const()[name = tensor<string, []>("op_10838_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10838_cast = slice_by_index(begin = var_10838_begin_0, end = var_10838_end_0, end_mask = var_10838_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10838_cast")];
            tensor<int32, [4]> var_10842_begin_0 = const()[name = tensor<string, []>("op_10842_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_10842_end_0 = const()[name = tensor<string, []>("op_10842_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_10842_end_mask_0 = const()[name = tensor<string, []>("op_10842_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_10842_cast = slice_by_index(begin = var_10842_begin_0, end = var_10842_end_0, end_mask = var_10842_end_mask_0, x = transpose_10)[name = tensor<string, []>("op_10842_cast")];
            tensor<int32, [4]> var_10844_begin_0 = const()[name = tensor<string, []>("op_10844_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_10844_end_0 = const()[name = tensor<string, []>("op_10844_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_10844_end_mask_0 = const()[name = tensor<string, []>("op_10844_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10844_cast = slice_by_index(begin = var_10844_begin_0, end = var_10844_end_0, end_mask = var_10844_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10844_cast")];
            tensor<int32, [4]> var_10848_begin_0 = const()[name = tensor<string, []>("op_10848_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_10848_end_0 = const()[name = tensor<string, []>("op_10848_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_10848_end_mask_0 = const()[name = tensor<string, []>("op_10848_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10848_cast = slice_by_index(begin = var_10848_begin_0, end = var_10848_end_0, end_mask = var_10848_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10848_cast")];
            tensor<int32, [4]> var_10852_begin_0 = const()[name = tensor<string, []>("op_10852_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_10852_end_0 = const()[name = tensor<string, []>("op_10852_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_10852_end_mask_0 = const()[name = tensor<string, []>("op_10852_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10852_cast = slice_by_index(begin = var_10852_begin_0, end = var_10852_end_0, end_mask = var_10852_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10852_cast")];
            tensor<int32, [4]> var_10856_begin_0 = const()[name = tensor<string, []>("op_10856_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_10856_end_0 = const()[name = tensor<string, []>("op_10856_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_10856_end_mask_0 = const()[name = tensor<string, []>("op_10856_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10856_cast = slice_by_index(begin = var_10856_begin_0, end = var_10856_end_0, end_mask = var_10856_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10856_cast")];
            tensor<int32, [4]> var_10860_begin_0 = const()[name = tensor<string, []>("op_10860_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_10860_end_0 = const()[name = tensor<string, []>("op_10860_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_10860_end_mask_0 = const()[name = tensor<string, []>("op_10860_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10860_cast = slice_by_index(begin = var_10860_begin_0, end = var_10860_end_0, end_mask = var_10860_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10860_cast")];
            tensor<int32, [4]> var_10864_begin_0 = const()[name = tensor<string, []>("op_10864_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_10864_end_0 = const()[name = tensor<string, []>("op_10864_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_10864_end_mask_0 = const()[name = tensor<string, []>("op_10864_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10864_cast = slice_by_index(begin = var_10864_begin_0, end = var_10864_end_0, end_mask = var_10864_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10864_cast")];
            tensor<int32, [4]> var_10868_begin_0 = const()[name = tensor<string, []>("op_10868_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_10868_end_0 = const()[name = tensor<string, []>("op_10868_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_10868_end_mask_0 = const()[name = tensor<string, []>("op_10868_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10868_cast = slice_by_index(begin = var_10868_begin_0, end = var_10868_end_0, end_mask = var_10868_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10868_cast")];
            tensor<int32, [4]> var_10872_begin_0 = const()[name = tensor<string, []>("op_10872_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_10872_end_0 = const()[name = tensor<string, []>("op_10872_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_10872_end_mask_0 = const()[name = tensor<string, []>("op_10872_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10872_cast = slice_by_index(begin = var_10872_begin_0, end = var_10872_end_0, end_mask = var_10872_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10872_cast")];
            tensor<int32, [4]> var_10876_begin_0 = const()[name = tensor<string, []>("op_10876_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_10876_end_0 = const()[name = tensor<string, []>("op_10876_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_10876_end_mask_0 = const()[name = tensor<string, []>("op_10876_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10876_cast = slice_by_index(begin = var_10876_begin_0, end = var_10876_end_0, end_mask = var_10876_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10876_cast")];
            tensor<int32, [4]> var_10880_begin_0 = const()[name = tensor<string, []>("op_10880_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_10880_end_0 = const()[name = tensor<string, []>("op_10880_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_10880_end_mask_0 = const()[name = tensor<string, []>("op_10880_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_10880_cast = slice_by_index(begin = var_10880_begin_0, end = var_10880_end_0, end_mask = var_10880_end_mask_0, x = v_43_cast)[name = tensor<string, []>("op_10880_cast")];
            tensor<string, []> var_10884_equation_0 = const()[name = tensor<string, []>("op_10884_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10884_cast = einsum(equation = var_10884_equation_0, values = (var_10806_cast, var_10782_cast))[name = tensor<string, []>("op_10884_cast")];
            tensor<fp16, []> var_10885_to_fp16 = const()[name = tensor<string, []>("op_10885_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_521_cast = mul(x = var_10884_cast, y = var_10885_to_fp16)[name = tensor<string, []>("aw_chunk_521_cast")];
            tensor<string, []> var_10888_equation_0 = const()[name = tensor<string, []>("op_10888_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10888_cast = einsum(equation = var_10888_equation_0, values = (var_10806_cast, var_10783_cast))[name = tensor<string, []>("op_10888_cast")];
            tensor<fp16, []> var_10889_to_fp16 = const()[name = tensor<string, []>("op_10889_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_523_cast = mul(x = var_10888_cast, y = var_10889_to_fp16)[name = tensor<string, []>("aw_chunk_523_cast")];
            tensor<string, []> var_10892_equation_0 = const()[name = tensor<string, []>("op_10892_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10892_cast = einsum(equation = var_10892_equation_0, values = (var_10810_cast, var_10784_cast))[name = tensor<string, []>("op_10892_cast")];
            tensor<fp16, []> var_10893_to_fp16 = const()[name = tensor<string, []>("op_10893_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_525_cast = mul(x = var_10892_cast, y = var_10893_to_fp16)[name = tensor<string, []>("aw_chunk_525_cast")];
            tensor<string, []> var_10896_equation_0 = const()[name = tensor<string, []>("op_10896_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10896_cast = einsum(equation = var_10896_equation_0, values = (var_10810_cast, var_10785_cast))[name = tensor<string, []>("op_10896_cast")];
            tensor<fp16, []> var_10897_to_fp16 = const()[name = tensor<string, []>("op_10897_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_527_cast = mul(x = var_10896_cast, y = var_10897_to_fp16)[name = tensor<string, []>("aw_chunk_527_cast")];
            tensor<string, []> var_10900_equation_0 = const()[name = tensor<string, []>("op_10900_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10900_cast = einsum(equation = var_10900_equation_0, values = (var_10814_cast, var_10786_cast))[name = tensor<string, []>("op_10900_cast")];
            tensor<fp16, []> var_10901_to_fp16 = const()[name = tensor<string, []>("op_10901_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_529_cast = mul(x = var_10900_cast, y = var_10901_to_fp16)[name = tensor<string, []>("aw_chunk_529_cast")];
            tensor<string, []> var_10904_equation_0 = const()[name = tensor<string, []>("op_10904_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10904_cast = einsum(equation = var_10904_equation_0, values = (var_10814_cast, var_10787_cast))[name = tensor<string, []>("op_10904_cast")];
            tensor<fp16, []> var_10905_to_fp16 = const()[name = tensor<string, []>("op_10905_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_531_cast = mul(x = var_10904_cast, y = var_10905_to_fp16)[name = tensor<string, []>("aw_chunk_531_cast")];
            tensor<string, []> var_10908_equation_0 = const()[name = tensor<string, []>("op_10908_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10908_cast = einsum(equation = var_10908_equation_0, values = (var_10818_cast, var_10788_cast))[name = tensor<string, []>("op_10908_cast")];
            tensor<fp16, []> var_10909_to_fp16 = const()[name = tensor<string, []>("op_10909_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_533_cast = mul(x = var_10908_cast, y = var_10909_to_fp16)[name = tensor<string, []>("aw_chunk_533_cast")];
            tensor<string, []> var_10912_equation_0 = const()[name = tensor<string, []>("op_10912_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10912_cast = einsum(equation = var_10912_equation_0, values = (var_10818_cast, var_10789_cast))[name = tensor<string, []>("op_10912_cast")];
            tensor<fp16, []> var_10913_to_fp16 = const()[name = tensor<string, []>("op_10913_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_535_cast = mul(x = var_10912_cast, y = var_10913_to_fp16)[name = tensor<string, []>("aw_chunk_535_cast")];
            tensor<string, []> var_10916_equation_0 = const()[name = tensor<string, []>("op_10916_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10916_cast = einsum(equation = var_10916_equation_0, values = (var_10822_cast, var_10790_cast))[name = tensor<string, []>("op_10916_cast")];
            tensor<fp16, []> var_10917_to_fp16 = const()[name = tensor<string, []>("op_10917_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_537_cast = mul(x = var_10916_cast, y = var_10917_to_fp16)[name = tensor<string, []>("aw_chunk_537_cast")];
            tensor<string, []> var_10920_equation_0 = const()[name = tensor<string, []>("op_10920_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10920_cast = einsum(equation = var_10920_equation_0, values = (var_10822_cast, var_10791_cast))[name = tensor<string, []>("op_10920_cast")];
            tensor<fp16, []> var_10921_to_fp16 = const()[name = tensor<string, []>("op_10921_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_539_cast = mul(x = var_10920_cast, y = var_10921_to_fp16)[name = tensor<string, []>("aw_chunk_539_cast")];
            tensor<string, []> var_10924_equation_0 = const()[name = tensor<string, []>("op_10924_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10924_cast = einsum(equation = var_10924_equation_0, values = (var_10826_cast, var_10792_cast))[name = tensor<string, []>("op_10924_cast")];
            tensor<fp16, []> var_10925_to_fp16 = const()[name = tensor<string, []>("op_10925_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_541_cast = mul(x = var_10924_cast, y = var_10925_to_fp16)[name = tensor<string, []>("aw_chunk_541_cast")];
            tensor<string, []> var_10928_equation_0 = const()[name = tensor<string, []>("op_10928_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10928_cast = einsum(equation = var_10928_equation_0, values = (var_10826_cast, var_10793_cast))[name = tensor<string, []>("op_10928_cast")];
            tensor<fp16, []> var_10929_to_fp16 = const()[name = tensor<string, []>("op_10929_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_543_cast = mul(x = var_10928_cast, y = var_10929_to_fp16)[name = tensor<string, []>("aw_chunk_543_cast")];
            tensor<string, []> var_10932_equation_0 = const()[name = tensor<string, []>("op_10932_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10932_cast = einsum(equation = var_10932_equation_0, values = (var_10830_cast, var_10794_cast))[name = tensor<string, []>("op_10932_cast")];
            tensor<fp16, []> var_10933_to_fp16 = const()[name = tensor<string, []>("op_10933_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_545_cast = mul(x = var_10932_cast, y = var_10933_to_fp16)[name = tensor<string, []>("aw_chunk_545_cast")];
            tensor<string, []> var_10936_equation_0 = const()[name = tensor<string, []>("op_10936_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10936_cast = einsum(equation = var_10936_equation_0, values = (var_10830_cast, var_10795_cast))[name = tensor<string, []>("op_10936_cast")];
            tensor<fp16, []> var_10937_to_fp16 = const()[name = tensor<string, []>("op_10937_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_547_cast = mul(x = var_10936_cast, y = var_10937_to_fp16)[name = tensor<string, []>("aw_chunk_547_cast")];
            tensor<string, []> var_10940_equation_0 = const()[name = tensor<string, []>("op_10940_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10940_cast = einsum(equation = var_10940_equation_0, values = (var_10834_cast, var_10796_cast))[name = tensor<string, []>("op_10940_cast")];
            tensor<fp16, []> var_10941_to_fp16 = const()[name = tensor<string, []>("op_10941_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_549_cast = mul(x = var_10940_cast, y = var_10941_to_fp16)[name = tensor<string, []>("aw_chunk_549_cast")];
            tensor<string, []> var_10944_equation_0 = const()[name = tensor<string, []>("op_10944_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10944_cast = einsum(equation = var_10944_equation_0, values = (var_10834_cast, var_10797_cast))[name = tensor<string, []>("op_10944_cast")];
            tensor<fp16, []> var_10945_to_fp16 = const()[name = tensor<string, []>("op_10945_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_551_cast = mul(x = var_10944_cast, y = var_10945_to_fp16)[name = tensor<string, []>("aw_chunk_551_cast")];
            tensor<string, []> var_10948_equation_0 = const()[name = tensor<string, []>("op_10948_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10948_cast = einsum(equation = var_10948_equation_0, values = (var_10838_cast, var_10798_cast))[name = tensor<string, []>("op_10948_cast")];
            tensor<fp16, []> var_10949_to_fp16 = const()[name = tensor<string, []>("op_10949_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_553_cast = mul(x = var_10948_cast, y = var_10949_to_fp16)[name = tensor<string, []>("aw_chunk_553_cast")];
            tensor<string, []> var_10952_equation_0 = const()[name = tensor<string, []>("op_10952_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10952_cast = einsum(equation = var_10952_equation_0, values = (var_10838_cast, var_10799_cast))[name = tensor<string, []>("op_10952_cast")];
            tensor<fp16, []> var_10953_to_fp16 = const()[name = tensor<string, []>("op_10953_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_555_cast = mul(x = var_10952_cast, y = var_10953_to_fp16)[name = tensor<string, []>("aw_chunk_555_cast")];
            tensor<string, []> var_10956_equation_0 = const()[name = tensor<string, []>("op_10956_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10956_cast = einsum(equation = var_10956_equation_0, values = (var_10842_cast, var_10800_cast))[name = tensor<string, []>("op_10956_cast")];
            tensor<fp16, []> var_10957_to_fp16 = const()[name = tensor<string, []>("op_10957_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_557_cast = mul(x = var_10956_cast, y = var_10957_to_fp16)[name = tensor<string, []>("aw_chunk_557_cast")];
            tensor<string, []> var_10960_equation_0 = const()[name = tensor<string, []>("op_10960_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_10960_cast = einsum(equation = var_10960_equation_0, values = (var_10842_cast, var_10801_cast))[name = tensor<string, []>("op_10960_cast")];
            tensor<fp16, []> var_10961_to_fp16 = const()[name = tensor<string, []>("op_10961_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_559_cast = mul(x = var_10960_cast, y = var_10961_to_fp16)[name = tensor<string, []>("aw_chunk_559_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10963_cast = softmax(axis = var_10257, x = aw_chunk_521_cast)[name = tensor<string, []>("op_10963_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10964_cast = softmax(axis = var_10257, x = aw_chunk_523_cast)[name = tensor<string, []>("op_10964_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10965_cast = softmax(axis = var_10257, x = aw_chunk_525_cast)[name = tensor<string, []>("op_10965_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10966_cast = softmax(axis = var_10257, x = aw_chunk_527_cast)[name = tensor<string, []>("op_10966_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10967_cast = softmax(axis = var_10257, x = aw_chunk_529_cast)[name = tensor<string, []>("op_10967_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10968_cast = softmax(axis = var_10257, x = aw_chunk_531_cast)[name = tensor<string, []>("op_10968_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10969_cast = softmax(axis = var_10257, x = aw_chunk_533_cast)[name = tensor<string, []>("op_10969_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10970_cast = softmax(axis = var_10257, x = aw_chunk_535_cast)[name = tensor<string, []>("op_10970_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10971_cast = softmax(axis = var_10257, x = aw_chunk_537_cast)[name = tensor<string, []>("op_10971_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10972_cast = softmax(axis = var_10257, x = aw_chunk_539_cast)[name = tensor<string, []>("op_10972_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10973_cast = softmax(axis = var_10257, x = aw_chunk_541_cast)[name = tensor<string, []>("op_10973_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10974_cast = softmax(axis = var_10257, x = aw_chunk_543_cast)[name = tensor<string, []>("op_10974_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10975_cast = softmax(axis = var_10257, x = aw_chunk_545_cast)[name = tensor<string, []>("op_10975_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10976_cast = softmax(axis = var_10257, x = aw_chunk_547_cast)[name = tensor<string, []>("op_10976_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10977_cast = softmax(axis = var_10257, x = aw_chunk_549_cast)[name = tensor<string, []>("op_10977_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10978_cast = softmax(axis = var_10257, x = aw_chunk_551_cast)[name = tensor<string, []>("op_10978_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10979_cast = softmax(axis = var_10257, x = aw_chunk_553_cast)[name = tensor<string, []>("op_10979_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10980_cast = softmax(axis = var_10257, x = aw_chunk_555_cast)[name = tensor<string, []>("op_10980_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10981_cast = softmax(axis = var_10257, x = aw_chunk_557_cast)[name = tensor<string, []>("op_10981_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_10982_cast = softmax(axis = var_10257, x = aw_chunk_559_cast)[name = tensor<string, []>("op_10982_cast")];
            tensor<string, []> var_10984_equation_0 = const()[name = tensor<string, []>("op_10984_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10984_cast = einsum(equation = var_10984_equation_0, values = (var_10844_cast, var_10963_cast))[name = tensor<string, []>("op_10984_cast")];
            tensor<string, []> var_10986_equation_0 = const()[name = tensor<string, []>("op_10986_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_10986_cast = einsum(equation = var_10986_equation_0, values = (var_10844_cast, var_10964_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, 64, 1, 512]> var_10988_cast = einsum(equation = var_10988_equation_0, values = (var_10848_cast, var_10965_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, 64, 1, 512]> var_10990_cast = einsum(equation = var_10990_equation_0, values = (var_10848_cast, var_10966_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, 64, 1, 512]> var_10992_cast = einsum(equation = var_10992_equation_0, values = (var_10852_cast, var_10967_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, 64, 1, 512]> var_10994_cast = einsum(equation = var_10994_equation_0, values = (var_10852_cast, var_10968_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, 64, 1, 512]> var_10996_cast = einsum(equation = var_10996_equation_0, values = (var_10856_cast, var_10969_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, 64, 1, 512]> var_10998_cast = einsum(equation = var_10998_equation_0, values = (var_10856_cast, var_10970_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, 64, 1, 512]> var_11000_cast = einsum(equation = var_11000_equation_0, values = (var_10860_cast, var_10971_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, 64, 1, 512]> var_11002_cast = einsum(equation = var_11002_equation_0, values = (var_10860_cast, var_10972_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, 64, 1, 512]> var_11004_cast = einsum(equation = var_11004_equation_0, values = (var_10864_cast, var_10973_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, 64, 1, 512]> var_11006_cast = einsum(equation = var_11006_equation_0, values = (var_10864_cast, var_10974_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, 64, 1, 512]> var_11008_cast = einsum(equation = var_11008_equation_0, values = (var_10868_cast, var_10975_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, 64, 1, 512]> var_11010_cast = einsum(equation = var_11010_equation_0, values = (var_10868_cast, var_10976_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, 64, 1, 512]> var_11012_cast = einsum(equation = var_11012_equation_0, values = (var_10872_cast, var_10977_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, 64, 1, 512]> var_11014_cast = einsum(equation = var_11014_equation_0, values = (var_10872_cast, var_10978_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, 64, 1, 512]> var_11016_cast = einsum(equation = var_11016_equation_0, values = (var_10876_cast, var_10979_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, 64, 1, 512]> var_11018_cast = einsum(equation = var_11018_equation_0, values = (var_10876_cast, var_10980_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, 64, 1, 512]> var_11020_cast = einsum(equation = var_11020_equation_0, values = (var_10880_cast, var_10981_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, 64, 1, 512]> var_11022_cast = einsum(equation = var_11022_equation_0, values = (var_10880_cast, var_10982_cast))[name = tensor<string, []>("op_11022_cast")];
            tensor<bool, []> var_11024_interleave_0 = const()[name = tensor<string, []>("op_11024_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11024_cast = concat(axis = var_10234, interleave = var_11024_interleave_0, values = (var_10984_cast, var_10986_cast))[name = tensor<string, []>("op_11024_cast")];
            tensor<bool, []> var_11026_interleave_0 = const()[name = tensor<string, []>("op_11026_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11026_cast = concat(axis = var_10234, interleave = var_11026_interleave_0, values = (var_10988_cast, var_10990_cast))[name = tensor<string, []>("op_11026_cast")];
            tensor<bool, []> var_11028_interleave_0 = const()[name = tensor<string, []>("op_11028_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11028_cast = concat(axis = var_10234, interleave = var_11028_interleave_0, values = (var_10992_cast, var_10994_cast))[name = tensor<string, []>("op_11028_cast")];
            tensor<bool, []> var_11030_interleave_0 = const()[name = tensor<string, []>("op_11030_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11030_cast = concat(axis = var_10234, interleave = var_11030_interleave_0, values = (var_10996_cast, var_10998_cast))[name = tensor<string, []>("op_11030_cast")];
            tensor<bool, []> var_11032_interleave_0 = const()[name = tensor<string, []>("op_11032_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11032_cast = concat(axis = var_10234, interleave = var_11032_interleave_0, values = (var_11000_cast, var_11002_cast))[name = tensor<string, []>("op_11032_cast")];
            tensor<bool, []> var_11034_interleave_0 = const()[name = tensor<string, []>("op_11034_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11034_cast = concat(axis = var_10234, interleave = var_11034_interleave_0, values = (var_11004_cast, var_11006_cast))[name = tensor<string, []>("op_11034_cast")];
            tensor<bool, []> var_11036_interleave_0 = const()[name = tensor<string, []>("op_11036_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11036_cast = concat(axis = var_10234, interleave = var_11036_interleave_0, values = (var_11008_cast, var_11010_cast))[name = tensor<string, []>("op_11036_cast")];
            tensor<bool, []> var_11038_interleave_0 = const()[name = tensor<string, []>("op_11038_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11038_cast = concat(axis = var_10234, interleave = var_11038_interleave_0, values = (var_11012_cast, var_11014_cast))[name = tensor<string, []>("op_11038_cast")];
            tensor<bool, []> var_11040_interleave_0 = const()[name = tensor<string, []>("op_11040_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11040_cast = concat(axis = var_10234, interleave = var_11040_interleave_0, values = (var_11016_cast, var_11018_cast))[name = tensor<string, []>("op_11040_cast")];
            tensor<bool, []> var_11042_interleave_0 = const()[name = tensor<string, []>("op_11042_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11042_cast = concat(axis = var_10234, interleave = var_11042_interleave_0, values = (var_11020_cast, var_11022_cast))[name = tensor<string, []>("op_11042_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_10257, interleave = input_385_interleave_0, values = (var_11024_cast, var_11026_cast, var_11028_cast, var_11030_cast, var_11032_cast, var_11034_cast, var_11036_cast, var_11038_cast, var_11040_cast, var_11042_cast))[name = tensor<string, []>("input_385_cast")];
            tensor<int32, [2]> var_11048 = const()[name = tensor<string, []>("op_11048"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11050 = const()[name = tensor<string, []>("op_11050"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11052_pad_type_0 = const()[name = tensor<string, []>("op_11052_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11052_pad_0 = const()[name = tensor<string, []>("op_11052_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199202688))), lut = tensor<fp16, [4]>([-0x1.48p-5, -0x1.82cp-7, 0x1.7a8p-7, 0x1.458p-5]), 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, []>(199305152)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11052_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11050, groups = var_10257, pad = var_11052_pad_0, pad_type = var_11052_pad_type_0, strides = var_11048, 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_11052_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_65_cast = add(x = var_11052_cast, y = inputs_63_cast)[name = tensor<string, []>("inputs_65_cast")];
            tensor<int32, [1]> var_11056 = const()[name = tensor<string, []>("op_11056"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_65_cast = reduce_mean(axes = var_11056, keep_dims = var_10252, 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_11060 = const()[name = tensor<string, []>("op_11060"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11061_cast = reduce_mean(axes = var_11060, keep_dims = var_10252, x = zero_mean_sq_65_cast)[name = tensor<string, []>("op_11061_cast")];
            tensor<fp16, []> var_11062_to_fp16 = const()[name = tensor<string, []>("op_11062_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11063_cast = add(x = var_11061_cast, y = var_11062_to_fp16)[name = tensor<string, []>("op_11063_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_11063_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_11067_to_fp16 = const()[name = tensor<string, []>("op_11067_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199306496)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11068_cast = add(x = out_65_cast, y = var_11067_to_fp16)[name = tensor<string, []>("op_11068_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_11070_to_fp16 = const()[name = tensor<string, []>("op_11070_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199307840)))];
            tensor<fp16, [2, 640, 1, 1024]> input_387_cast = mul(x = var_11068_cast, y = var_11070_to_fp16)[name = tensor<string, []>("input_387_cast")];
            tensor<int32, [2]> var_11078 = const()[name = tensor<string, []>("op_11078"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11080 = const()[name = tensor<string, []>("op_11080"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11082_pad_type_0 = const()[name = tensor<string, []>("op_11082_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11082_pad_0 = const()[name = tensor<string, []>("op_11082_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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199309184))), lut = tensor<fp16, [4]>([-0x1.9ecp-4, -0x1.e6p-6, 0x1.d9p-6, 0x1.988p-4]), 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200128448))), lut = tensor<fp16, [4]>([0x1.0dcp-4, -0x1.658p-4, -0x1.cep-9, -0x1.2b4p-2]), 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_11082_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_11080, groups = var_10257, pad = var_11082_pad_0, pad_type = var_11082_pad_type_0, strides = var_11078, 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_11082_cast")];
            tensor<int32, [2]> var_11083_split_sizes_0 = const()[name = tensor<string, []>("op_11083_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_11083_axis_0 = const()[name = tensor<string, []>("op_11083_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_11083_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_11083_cast_1 = split(axis = var_11083_axis_0, split_sizes = var_11083_split_sizes_0, x = var_11082_cast)[name = tensor<string, []>("op_11083_cast")];
            tensor<string, []> var_11085_mode_0 = const()[name = tensor<string, []>("op_11085_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_11085_cast = gelu(mode = var_11085_mode_0, x = var_11083_cast_1)[name = tensor<string, []>("op_11085_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_389_cast = mul(x = var_11083_cast_0, y = var_11085_cast)[name = tensor<string, []>("input_389_cast")];
            tensor<int32, [2]> var_11089 = const()[name = tensor<string, []>("op_11089"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11091 = const()[name = tensor<string, []>("op_11091"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11093_pad_type_0 = const()[name = tensor<string, []>("op_11093_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11093_pad_0 = const()[name = tensor<string, []>("op_11093_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200129792))), lut = tensor<fp16, [4]>([-0x1.abcp-4, -0x1.fbcp-6, 0x1.fep-6, 0x1.ac4p-4]), 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, []>(200539456)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11093_cast = conv(bias = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11091, groups = var_10257, pad = var_11093_pad_0, pad_type = var_11093_pad_type_0, strides = var_11089, weight = up_blocks_2_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_389_cast)[name = tensor<string, []>("op_11093_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_233_cast = add(x = var_11093_cast, y = inputs_65_cast)[name = tensor<string, []>("hidden_states_233_cast")];
            tensor<int32, [4]> var_11095 = const()[name = tensor<string, []>("op_11095"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_391_cast = reshape(shape = var_11095, x = hidden_states_233_cast)[name = tensor<string, []>("input_391_cast")];
            tensor<int32, [2]> var_11099 = const()[name = tensor<string, []>("op_11099"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11101 = const()[name = tensor<string, []>("op_11101"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200540800))), lut = tensor<fp16, [4]>([-0x1.2c8p-3, -0x1.668p-5, 0x1.6a4p-5, 0x1.2e4p-3]), 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, []>(200643264)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_235_cast = conv(bias = up_blocks_2_attentions_0_proj_out_bias_to_fp16, dilations = var_11101, groups = var_10257, pad = hidden_states_235_pad_0, pad_type = hidden_states_235_pad_type_0, strides = var_11099, 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_10257, 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, []>(200644608)))];
            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, []>(200647232)))];
            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_11119 = const()[name = tensor<string, []>("op_11119"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11121 = const()[name = tensor<string, []>("op_11121"), 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, [1843200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200649856))), lut = tensor<fp16, [4]>([-0x1.988p-4, -0x1.84cp-6, 0x1.a2p-6, 0x1.aacp-4]), 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, []>(202493120)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_239_cast = conv(bias = up_blocks_2_resnets_1_conv1_bias_to_fp16, dilations = var_11121, groups = var_10257, pad = hidden_states_239_pad_0, pad_type = hidden_states_239_pad_type_0, strides = var_11119, 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_11127 = const()[name = tensor<string, []>("op_11127"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11129 = const()[name = tensor<string, []>("op_11129"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202494464))), lut = tensor<fp16, [4]>([-0x1.878p-5, -0x1.7b8p-8, 0x1.8c8p-11, 0x1.038p-7]), 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, []>(202699328)))];
            tensor<fp16, [2, 640, 1, 1]> temb_35_cast = conv(bias = up_blocks_2_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_11129, groups = var_10257, pad = temb_35_pad_0, pad_type = temb_35_pad_type_0, strides = var_11127, 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, []>(202700672)))];
            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, []>(202702016)))];
            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_11139 = const()[name = tensor<string, []>("op_11139"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11141 = const()[name = tensor<string, []>("op_11141"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202703360))), lut = tensor<fp16, [4]>([-0x1.e08p-4, -0x1.b3p-6, 0x1.aecp-6, 0x1.dep-4]), 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, []>(203625024)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_241_cast = conv(bias = up_blocks_2_resnets_1_conv2_bias_to_fp16, dilations = var_11141, groups = var_10257, pad = hidden_states_241_pad_0, pad_type = hidden_states_241_pad_type_0, strides = var_11139, 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_11146 = const()[name = tensor<string, []>("op_11146"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11148 = const()[name = tensor<string, []>("op_11148"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203626368))), lut = tensor<fp16, [4]>([-0x1.b3p-5, -0x1.9ecp-7, 0x1.b28p-7, 0x1.bap-5]), 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, []>(203831232)))];
            tensor<fp16, [2, 640, 32, 32]> x_19_cast = conv(bias = up_blocks_2_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_11148, groups = var_10257, pad = x_19_pad_0, pad_type = x_19_pad_type_0, strides = var_11146, 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, []>(203832576)))];
            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, []>(203833920)))];
            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_11168 = const()[name = tensor<string, []>("op_11168"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11170 = const()[name = tensor<string, []>("op_11170"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203835264))), lut = tensor<fp16, [4]>([-0x1.a28p-4, -0x1.e98p-6, 0x1.01p-5, 0x1.aacp-4]), 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, []>(203937728)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_245_cast = conv(bias = up_blocks_2_attentions_1_proj_in_bias_to_fp16, dilations = var_11170, groups = var_10257, pad = hidden_states_245_pad_0, pad_type = hidden_states_245_pad_type_0, strides = var_11168, 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_11175 = const()[name = tensor<string, []>("op_11175"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_67_cast = reshape(shape = var_11175, x = hidden_states_245_cast)[name = tensor<string, []>("inputs_67_cast")];
            tensor<int32, [1]> var_11185 = const()[name = tensor<string, []>("op_11185"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_67_cast = reduce_mean(axes = var_11185, keep_dims = var_10252, 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_11189 = const()[name = tensor<string, []>("op_11189"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11190_cast = reduce_mean(axes = var_11189, keep_dims = var_10252, x = zero_mean_sq_67_cast)[name = tensor<string, []>("op_11190_cast")];
            tensor<fp16, []> var_11191_to_fp16 = const()[name = tensor<string, []>("op_11191_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11192_cast = add(x = var_11190_cast, y = var_11191_to_fp16)[name = tensor<string, []>("op_11192_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_11192_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_11196_to_fp16 = const()[name = tensor<string, []>("op_11196_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203939072)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11197_cast = add(x = out_67_cast, y = var_11196_to_fp16)[name = tensor<string, []>("op_11197_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_11199_to_fp16 = const()[name = tensor<string, []>("op_11199_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203940416)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_247_cast = mul(x = var_11197_cast, y = var_11199_to_fp16)[name = tensor<string, []>("hidden_states_247_cast")];
            tensor<int32, [2]> var_11206 = const()[name = tensor<string, []>("op_11206"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11208 = const()[name = tensor<string, []>("op_11208"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203941760))), lut = tensor<fp16, [4]>([-0x1.e54p-4, -0x1.1bp-5, 0x1.118p-5, 0x1.e18p-4]), 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_11208, groups = var_10257, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = var_11206, 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_11212 = const()[name = tensor<string, []>("op_11212"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11214 = const()[name = tensor<string, []>("op_11214"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204044224))), lut = tensor<fp16, [4]>([-0x1.d44p-4, -0x1.0ep-5, 0x1.0b4p-5, 0x1.d28p-4]), 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_11214, groups = var_10257, pad = k_89_pad_0, pad_type = k_89_pad_type_0, strides = var_11212, 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_11218 = const()[name = tensor<string, []>("op_11218"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11220 = const()[name = tensor<string, []>("op_11220"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204146688))), lut = tensor<fp16, [4]>([-0x1.a38p-4, -0x1.e7p-6, 0x1.e8cp-6, 0x1.a38p-4]), 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_11220, groups = var_10257, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = var_11218, 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_11224_begin_0 = const()[name = tensor<string, []>("op_11224_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11224_end_0 = const()[name = tensor<string, []>("op_11224_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11224_end_mask_0 = const()[name = tensor<string, []>("op_11224_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11224_cast = slice_by_index(begin = var_11224_begin_0, end = var_11224_end_0, end_mask = var_11224_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11224_cast")];
            tensor<int32, [4]> var_11228_begin_0 = const()[name = tensor<string, []>("op_11228_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11228_end_0 = const()[name = tensor<string, []>("op_11228_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_11228_end_mask_0 = const()[name = tensor<string, []>("op_11228_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11228_cast = slice_by_index(begin = var_11228_begin_0, end = var_11228_end_0, end_mask = var_11228_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11228_cast")];
            tensor<int32, [4]> var_11232_begin_0 = const()[name = tensor<string, []>("op_11232_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11232_end_0 = const()[name = tensor<string, []>("op_11232_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_11232_end_mask_0 = const()[name = tensor<string, []>("op_11232_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11232_cast = slice_by_index(begin = var_11232_begin_0, end = var_11232_end_0, end_mask = var_11232_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11232_cast")];
            tensor<int32, [4]> var_11236_begin_0 = const()[name = tensor<string, []>("op_11236_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11236_end_0 = const()[name = tensor<string, []>("op_11236_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_11236_end_mask_0 = const()[name = tensor<string, []>("op_11236_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11236_cast = slice_by_index(begin = var_11236_begin_0, end = var_11236_end_0, end_mask = var_11236_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11236_cast")];
            tensor<int32, [4]> var_11240_begin_0 = const()[name = tensor<string, []>("op_11240_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11240_end_0 = const()[name = tensor<string, []>("op_11240_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_11240_end_mask_0 = const()[name = tensor<string, []>("op_11240_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11240_cast = slice_by_index(begin = var_11240_begin_0, end = var_11240_end_0, end_mask = var_11240_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11240_cast")];
            tensor<int32, [4]> var_11244_begin_0 = const()[name = tensor<string, []>("op_11244_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11244_end_0 = const()[name = tensor<string, []>("op_11244_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_11244_end_mask_0 = const()[name = tensor<string, []>("op_11244_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11244_cast = slice_by_index(begin = var_11244_begin_0, end = var_11244_end_0, end_mask = var_11244_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11244_cast")];
            tensor<int32, [4]> var_11248_begin_0 = const()[name = tensor<string, []>("op_11248_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11248_end_0 = const()[name = tensor<string, []>("op_11248_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_11248_end_mask_0 = const()[name = tensor<string, []>("op_11248_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11248_cast = slice_by_index(begin = var_11248_begin_0, end = var_11248_end_0, end_mask = var_11248_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11248_cast")];
            tensor<int32, [4]> var_11252_begin_0 = const()[name = tensor<string, []>("op_11252_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11252_end_0 = const()[name = tensor<string, []>("op_11252_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_11252_end_mask_0 = const()[name = tensor<string, []>("op_11252_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11252_cast = slice_by_index(begin = var_11252_begin_0, end = var_11252_end_0, end_mask = var_11252_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11252_cast")];
            tensor<int32, [4]> var_11256_begin_0 = const()[name = tensor<string, []>("op_11256_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11256_end_0 = const()[name = tensor<string, []>("op_11256_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_11256_end_mask_0 = const()[name = tensor<string, []>("op_11256_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11256_cast = slice_by_index(begin = var_11256_begin_0, end = var_11256_end_0, end_mask = var_11256_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11256_cast")];
            tensor<int32, [4]> var_11260_begin_0 = const()[name = tensor<string, []>("op_11260_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11260_end_0 = const()[name = tensor<string, []>("op_11260_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_11260_end_mask_0 = const()[name = tensor<string, []>("op_11260_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11260_cast = slice_by_index(begin = var_11260_begin_0, end = var_11260_end_0, end_mask = var_11260_end_mask_0, x = q_45_cast)[name = tensor<string, []>("op_11260_cast")];
            tensor<int32, [4]> var_11263_begin_0 = const()[name = tensor<string, []>("op_11263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11263_end_0 = const()[name = tensor<string, []>("op_11263_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11224_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, 512])];
            tensor<int32, [4]> var_11264_end_0 = const()[name = tensor<string, []>("op_11264_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11224_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, 0])];
            tensor<int32, [4]> var_11265_end_0 = const()[name = tensor<string, []>("op_11265_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11228_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, 512])];
            tensor<int32, [4]> var_11266_end_0 = const()[name = tensor<string, []>("op_11266_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11228_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, 0])];
            tensor<int32, [4]> var_11267_end_0 = const()[name = tensor<string, []>("op_11267_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11232_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, 512])];
            tensor<int32, [4]> var_11268_end_0 = const()[name = tensor<string, []>("op_11268_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11232_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, 0])];
            tensor<int32, [4]> var_11269_end_0 = const()[name = tensor<string, []>("op_11269_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11236_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, 512])];
            tensor<int32, [4]> var_11270_end_0 = const()[name = tensor<string, []>("op_11270_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11236_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, 0])];
            tensor<int32, [4]> var_11271_end_0 = const()[name = tensor<string, []>("op_11271_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11240_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, 512])];
            tensor<int32, [4]> var_11272_end_0 = const()[name = tensor<string, []>("op_11272_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11240_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, 0])];
            tensor<int32, [4]> var_11273_end_0 = const()[name = tensor<string, []>("op_11273_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11244_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, 512])];
            tensor<int32, [4]> var_11274_end_0 = const()[name = tensor<string, []>("op_11274_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11244_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, 0])];
            tensor<int32, [4]> var_11275_end_0 = const()[name = tensor<string, []>("op_11275_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11248_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, 512])];
            tensor<int32, [4]> var_11276_end_0 = const()[name = tensor<string, []>("op_11276_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 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_11248_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, 0])];
            tensor<int32, [4]> var_11277_end_0 = const()[name = tensor<string, []>("op_11277_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            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, 64, 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_11252_cast)[name = tensor<string, []>("op_11277_cast")];
            tensor<int32, [4]> var_11278_begin_0 = const()[name = tensor<string, []>("op_11278_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11278_end_0 = const()[name = tensor<string, []>("op_11278_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11278_end_mask_0 = const()[name = tensor<string, []>("op_11278_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11278_cast = slice_by_index(begin = var_11278_begin_0, end = var_11278_end_0, end_mask = var_11278_end_mask_0, x = var_11252_cast)[name = tensor<string, []>("op_11278_cast")];
            tensor<int32, [4]> var_11279_begin_0 = const()[name = tensor<string, []>("op_11279_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11279_end_0 = const()[name = tensor<string, []>("op_11279_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11279_end_mask_0 = const()[name = tensor<string, []>("op_11279_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11279_cast = slice_by_index(begin = var_11279_begin_0, end = var_11279_end_0, end_mask = var_11279_end_mask_0, x = var_11256_cast)[name = tensor<string, []>("op_11279_cast")];
            tensor<int32, [4]> var_11280_begin_0 = const()[name = tensor<string, []>("op_11280_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11280_end_0 = const()[name = tensor<string, []>("op_11280_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11280_end_mask_0 = const()[name = tensor<string, []>("op_11280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11280_cast = slice_by_index(begin = var_11280_begin_0, end = var_11280_end_0, end_mask = var_11280_end_mask_0, x = var_11256_cast)[name = tensor<string, []>("op_11280_cast")];
            tensor<int32, [4]> var_11281_begin_0 = const()[name = tensor<string, []>("op_11281_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11281_end_0 = const()[name = tensor<string, []>("op_11281_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11281_end_mask_0 = const()[name = tensor<string, []>("op_11281_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11281_cast = slice_by_index(begin = var_11281_begin_0, end = var_11281_end_0, end_mask = var_11281_end_mask_0, x = var_11260_cast)[name = tensor<string, []>("op_11281_cast")];
            tensor<int32, [4]> var_11282_begin_0 = const()[name = tensor<string, []>("op_11282_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11282_end_0 = const()[name = tensor<string, []>("op_11282_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 1, 512]> var_11282_cast = slice_by_index(begin = var_11282_begin_0, end = var_11282_end_0, end_mask = var_11282_end_mask_0, x = var_11260_cast)[name = tensor<string, []>("op_11282_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_11287_begin_0 = const()[name = tensor<string, []>("op_11287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11287_end_0 = const()[name = tensor<string, []>("op_11287_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_11287_end_mask_0 = const()[name = tensor<string, []>("op_11287_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, 64]> var_11287_cast = slice_by_index(begin = var_11287_begin_0, end = var_11287_end_0, end_mask = var_11287_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11287_cast")];
            tensor<int32, [4]> var_11291_begin_0 = const()[name = tensor<string, []>("op_11291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11291_end_0 = const()[name = tensor<string, []>("op_11291_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_11291_end_mask_0 = const()[name = tensor<string, []>("op_11291_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11291_cast = slice_by_index(begin = var_11291_begin_0, end = var_11291_end_0, end_mask = var_11291_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11291_cast")];
            tensor<int32, [4]> var_11295_begin_0 = const()[name = tensor<string, []>("op_11295_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11295_end_0 = const()[name = tensor<string, []>("op_11295_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_11295_end_mask_0 = const()[name = tensor<string, []>("op_11295_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11295_cast = slice_by_index(begin = var_11295_begin_0, end = var_11295_end_0, end_mask = var_11295_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11295_cast")];
            tensor<int32, [4]> var_11299_begin_0 = const()[name = tensor<string, []>("op_11299_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11299_end_0 = const()[name = tensor<string, []>("op_11299_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_11299_end_mask_0 = const()[name = tensor<string, []>("op_11299_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11299_cast = slice_by_index(begin = var_11299_begin_0, end = var_11299_end_0, end_mask = var_11299_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11299_cast")];
            tensor<int32, [4]> var_11303_begin_0 = const()[name = tensor<string, []>("op_11303_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11303_end_0 = const()[name = tensor<string, []>("op_11303_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_11303_end_mask_0 = const()[name = tensor<string, []>("op_11303_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11303_cast = slice_by_index(begin = var_11303_begin_0, end = var_11303_end_0, end_mask = var_11303_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11303_cast")];
            tensor<int32, [4]> var_11307_begin_0 = const()[name = tensor<string, []>("op_11307_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_11307_end_0 = const()[name = tensor<string, []>("op_11307_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_11307_end_mask_0 = const()[name = tensor<string, []>("op_11307_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11307_cast = slice_by_index(begin = var_11307_begin_0, end = var_11307_end_0, end_mask = var_11307_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11307_cast")];
            tensor<int32, [4]> var_11311_begin_0 = const()[name = tensor<string, []>("op_11311_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_11311_end_0 = const()[name = tensor<string, []>("op_11311_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_11311_end_mask_0 = const()[name = tensor<string, []>("op_11311_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11311_cast = slice_by_index(begin = var_11311_begin_0, end = var_11311_end_0, end_mask = var_11311_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11311_cast")];
            tensor<int32, [4]> var_11315_begin_0 = const()[name = tensor<string, []>("op_11315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_11315_end_0 = const()[name = tensor<string, []>("op_11315_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_11315_end_mask_0 = const()[name = tensor<string, []>("op_11315_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11315_cast = slice_by_index(begin = var_11315_begin_0, end = var_11315_end_0, end_mask = var_11315_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11315_cast")];
            tensor<int32, [4]> var_11319_begin_0 = const()[name = tensor<string, []>("op_11319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11319_end_0 = const()[name = tensor<string, []>("op_11319_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_11319_end_mask_0 = const()[name = tensor<string, []>("op_11319_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11319_cast = slice_by_index(begin = var_11319_begin_0, end = var_11319_end_0, end_mask = var_11319_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11319_cast")];
            tensor<int32, [4]> var_11323_begin_0 = const()[name = tensor<string, []>("op_11323_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_11323_end_0 = const()[name = tensor<string, []>("op_11323_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_11323_end_mask_0 = const()[name = tensor<string, []>("op_11323_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_11323_cast = slice_by_index(begin = var_11323_begin_0, end = var_11323_end_0, end_mask = var_11323_end_mask_0, x = transpose_9)[name = tensor<string, []>("op_11323_cast")];
            tensor<int32, [4]> var_11325_begin_0 = const()[name = tensor<string, []>("op_11325_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11325_end_0 = const()[name = tensor<string, []>("op_11325_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11325_end_mask_0 = const()[name = tensor<string, []>("op_11325_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11325_cast = slice_by_index(begin = var_11325_begin_0, end = var_11325_end_0, end_mask = var_11325_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11325_cast")];
            tensor<int32, [4]> var_11329_begin_0 = const()[name = tensor<string, []>("op_11329_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11329_end_0 = const()[name = tensor<string, []>("op_11329_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_11329_end_mask_0 = const()[name = tensor<string, []>("op_11329_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11329_cast = slice_by_index(begin = var_11329_begin_0, end = var_11329_end_0, end_mask = var_11329_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11329_cast")];
            tensor<int32, [4]> var_11333_begin_0 = const()[name = tensor<string, []>("op_11333_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11333_end_0 = const()[name = tensor<string, []>("op_11333_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_11333_end_mask_0 = const()[name = tensor<string, []>("op_11333_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11333_cast = slice_by_index(begin = var_11333_begin_0, end = var_11333_end_0, end_mask = var_11333_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11333_cast")];
            tensor<int32, [4]> var_11337_begin_0 = const()[name = tensor<string, []>("op_11337_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11337_end_0 = const()[name = tensor<string, []>("op_11337_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_11337_end_mask_0 = const()[name = tensor<string, []>("op_11337_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11337_cast = slice_by_index(begin = var_11337_begin_0, end = var_11337_end_0, end_mask = var_11337_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11337_cast")];
            tensor<int32, [4]> var_11341_begin_0 = const()[name = tensor<string, []>("op_11341_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11341_end_0 = const()[name = tensor<string, []>("op_11341_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_11341_end_mask_0 = const()[name = tensor<string, []>("op_11341_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11341_cast = slice_by_index(begin = var_11341_begin_0, end = var_11341_end_0, end_mask = var_11341_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11341_cast")];
            tensor<int32, [4]> var_11345_begin_0 = const()[name = tensor<string, []>("op_11345_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11345_end_0 = const()[name = tensor<string, []>("op_11345_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_11345_end_mask_0 = const()[name = tensor<string, []>("op_11345_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11345_cast = slice_by_index(begin = var_11345_begin_0, end = var_11345_end_0, end_mask = var_11345_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11345_cast")];
            tensor<int32, [4]> var_11349_begin_0 = const()[name = tensor<string, []>("op_11349_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11349_end_0 = const()[name = tensor<string, []>("op_11349_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_11349_end_mask_0 = const()[name = tensor<string, []>("op_11349_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11349_cast = slice_by_index(begin = var_11349_begin_0, end = var_11349_end_0, end_mask = var_11349_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11349_cast")];
            tensor<int32, [4]> var_11353_begin_0 = const()[name = tensor<string, []>("op_11353_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11353_end_0 = const()[name = tensor<string, []>("op_11353_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_11353_end_mask_0 = const()[name = tensor<string, []>("op_11353_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11353_cast = slice_by_index(begin = var_11353_begin_0, end = var_11353_end_0, end_mask = var_11353_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11353_cast")];
            tensor<int32, [4]> var_11357_begin_0 = const()[name = tensor<string, []>("op_11357_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11357_end_0 = const()[name = tensor<string, []>("op_11357_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_11357_end_mask_0 = const()[name = tensor<string, []>("op_11357_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11357_cast = slice_by_index(begin = var_11357_begin_0, end = var_11357_end_0, end_mask = var_11357_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11357_cast")];
            tensor<int32, [4]> var_11361_begin_0 = const()[name = tensor<string, []>("op_11361_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11361_end_0 = const()[name = tensor<string, []>("op_11361_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_11361_end_mask_0 = const()[name = tensor<string, []>("op_11361_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11361_cast = slice_by_index(begin = var_11361_begin_0, end = var_11361_end_0, end_mask = var_11361_end_mask_0, x = v_45_cast)[name = tensor<string, []>("op_11361_cast")];
            tensor<string, []> var_11365_equation_0 = const()[name = tensor<string, []>("op_11365_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11365_cast = einsum(equation = var_11365_equation_0, values = (var_11287_cast, var_11263_cast))[name = tensor<string, []>("op_11365_cast")];
            tensor<fp16, []> var_11366_to_fp16 = const()[name = tensor<string, []>("op_11366_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_561_cast = mul(x = var_11365_cast, y = var_11366_to_fp16)[name = tensor<string, []>("aw_chunk_561_cast")];
            tensor<string, []> var_11369_equation_0 = const()[name = tensor<string, []>("op_11369_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11369_cast = einsum(equation = var_11369_equation_0, values = (var_11287_cast, var_11264_cast))[name = tensor<string, []>("op_11369_cast")];
            tensor<fp16, []> var_11370_to_fp16 = const()[name = tensor<string, []>("op_11370_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_563_cast = mul(x = var_11369_cast, y = var_11370_to_fp16)[name = tensor<string, []>("aw_chunk_563_cast")];
            tensor<string, []> var_11373_equation_0 = const()[name = tensor<string, []>("op_11373_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11373_cast = einsum(equation = var_11373_equation_0, values = (var_11291_cast, var_11265_cast))[name = tensor<string, []>("op_11373_cast")];
            tensor<fp16, []> var_11374_to_fp16 = const()[name = tensor<string, []>("op_11374_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_565_cast = mul(x = var_11373_cast, y = var_11374_to_fp16)[name = tensor<string, []>("aw_chunk_565_cast")];
            tensor<string, []> var_11377_equation_0 = const()[name = tensor<string, []>("op_11377_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11377_cast = einsum(equation = var_11377_equation_0, values = (var_11291_cast, var_11266_cast))[name = tensor<string, []>("op_11377_cast")];
            tensor<fp16, []> var_11378_to_fp16 = const()[name = tensor<string, []>("op_11378_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_567_cast = mul(x = var_11377_cast, y = var_11378_to_fp16)[name = tensor<string, []>("aw_chunk_567_cast")];
            tensor<string, []> var_11381_equation_0 = const()[name = tensor<string, []>("op_11381_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11381_cast = einsum(equation = var_11381_equation_0, values = (var_11295_cast, var_11267_cast))[name = tensor<string, []>("op_11381_cast")];
            tensor<fp16, []> var_11382_to_fp16 = const()[name = tensor<string, []>("op_11382_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_569_cast = mul(x = var_11381_cast, y = var_11382_to_fp16)[name = tensor<string, []>("aw_chunk_569_cast")];
            tensor<string, []> var_11385_equation_0 = const()[name = tensor<string, []>("op_11385_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11385_cast = einsum(equation = var_11385_equation_0, values = (var_11295_cast, var_11268_cast))[name = tensor<string, []>("op_11385_cast")];
            tensor<fp16, []> var_11386_to_fp16 = const()[name = tensor<string, []>("op_11386_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_571_cast = mul(x = var_11385_cast, y = var_11386_to_fp16)[name = tensor<string, []>("aw_chunk_571_cast")];
            tensor<string, []> var_11389_equation_0 = const()[name = tensor<string, []>("op_11389_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11389_cast = einsum(equation = var_11389_equation_0, values = (var_11299_cast, var_11269_cast))[name = tensor<string, []>("op_11389_cast")];
            tensor<fp16, []> var_11390_to_fp16 = const()[name = tensor<string, []>("op_11390_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_573_cast = mul(x = var_11389_cast, y = var_11390_to_fp16)[name = tensor<string, []>("aw_chunk_573_cast")];
            tensor<string, []> var_11393_equation_0 = const()[name = tensor<string, []>("op_11393_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11393_cast = einsum(equation = var_11393_equation_0, values = (var_11299_cast, var_11270_cast))[name = tensor<string, []>("op_11393_cast")];
            tensor<fp16, []> var_11394_to_fp16 = const()[name = tensor<string, []>("op_11394_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_575_cast = mul(x = var_11393_cast, y = var_11394_to_fp16)[name = tensor<string, []>("aw_chunk_575_cast")];
            tensor<string, []> var_11397_equation_0 = const()[name = tensor<string, []>("op_11397_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11397_cast = einsum(equation = var_11397_equation_0, values = (var_11303_cast, var_11271_cast))[name = tensor<string, []>("op_11397_cast")];
            tensor<fp16, []> var_11398_to_fp16 = const()[name = tensor<string, []>("op_11398_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_577_cast = mul(x = var_11397_cast, y = var_11398_to_fp16)[name = tensor<string, []>("aw_chunk_577_cast")];
            tensor<string, []> var_11401_equation_0 = const()[name = tensor<string, []>("op_11401_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11401_cast = einsum(equation = var_11401_equation_0, values = (var_11303_cast, var_11272_cast))[name = tensor<string, []>("op_11401_cast")];
            tensor<fp16, []> var_11402_to_fp16 = const()[name = tensor<string, []>("op_11402_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_579_cast = mul(x = var_11401_cast, y = var_11402_to_fp16)[name = tensor<string, []>("aw_chunk_579_cast")];
            tensor<string, []> var_11405_equation_0 = const()[name = tensor<string, []>("op_11405_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11405_cast = einsum(equation = var_11405_equation_0, values = (var_11307_cast, var_11273_cast))[name = tensor<string, []>("op_11405_cast")];
            tensor<fp16, []> var_11406_to_fp16 = const()[name = tensor<string, []>("op_11406_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_581_cast = mul(x = var_11405_cast, y = var_11406_to_fp16)[name = tensor<string, []>("aw_chunk_581_cast")];
            tensor<string, []> var_11409_equation_0 = const()[name = tensor<string, []>("op_11409_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11409_cast = einsum(equation = var_11409_equation_0, values = (var_11307_cast, var_11274_cast))[name = tensor<string, []>("op_11409_cast")];
            tensor<fp16, []> var_11410_to_fp16 = const()[name = tensor<string, []>("op_11410_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_583_cast = mul(x = var_11409_cast, y = var_11410_to_fp16)[name = tensor<string, []>("aw_chunk_583_cast")];
            tensor<string, []> var_11413_equation_0 = const()[name = tensor<string, []>("op_11413_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11413_cast = einsum(equation = var_11413_equation_0, values = (var_11311_cast, var_11275_cast))[name = tensor<string, []>("op_11413_cast")];
            tensor<fp16, []> var_11414_to_fp16 = const()[name = tensor<string, []>("op_11414_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_585_cast = mul(x = var_11413_cast, y = var_11414_to_fp16)[name = tensor<string, []>("aw_chunk_585_cast")];
            tensor<string, []> var_11417_equation_0 = const()[name = tensor<string, []>("op_11417_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11417_cast = einsum(equation = var_11417_equation_0, values = (var_11311_cast, var_11276_cast))[name = tensor<string, []>("op_11417_cast")];
            tensor<fp16, []> var_11418_to_fp16 = const()[name = tensor<string, []>("op_11418_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_587_cast = mul(x = var_11417_cast, y = var_11418_to_fp16)[name = tensor<string, []>("aw_chunk_587_cast")];
            tensor<string, []> var_11421_equation_0 = const()[name = tensor<string, []>("op_11421_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11421_cast = einsum(equation = var_11421_equation_0, values = (var_11315_cast, var_11277_cast))[name = tensor<string, []>("op_11421_cast")];
            tensor<fp16, []> var_11422_to_fp16 = const()[name = tensor<string, []>("op_11422_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_589_cast = mul(x = var_11421_cast, y = var_11422_to_fp16)[name = tensor<string, []>("aw_chunk_589_cast")];
            tensor<string, []> var_11425_equation_0 = const()[name = tensor<string, []>("op_11425_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11425_cast = einsum(equation = var_11425_equation_0, values = (var_11315_cast, var_11278_cast))[name = tensor<string, []>("op_11425_cast")];
            tensor<fp16, []> var_11426_to_fp16 = const()[name = tensor<string, []>("op_11426_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_591_cast = mul(x = var_11425_cast, y = var_11426_to_fp16)[name = tensor<string, []>("aw_chunk_591_cast")];
            tensor<string, []> var_11429_equation_0 = const()[name = tensor<string, []>("op_11429_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11429_cast = einsum(equation = var_11429_equation_0, values = (var_11319_cast, var_11279_cast))[name = tensor<string, []>("op_11429_cast")];
            tensor<fp16, []> var_11430_to_fp16 = const()[name = tensor<string, []>("op_11430_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_593_cast = mul(x = var_11429_cast, y = var_11430_to_fp16)[name = tensor<string, []>("aw_chunk_593_cast")];
            tensor<string, []> var_11433_equation_0 = const()[name = tensor<string, []>("op_11433_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11433_cast = einsum(equation = var_11433_equation_0, values = (var_11319_cast, var_11280_cast))[name = tensor<string, []>("op_11433_cast")];
            tensor<fp16, []> var_11434_to_fp16 = const()[name = tensor<string, []>("op_11434_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_595_cast = mul(x = var_11433_cast, y = var_11434_to_fp16)[name = tensor<string, []>("aw_chunk_595_cast")];
            tensor<string, []> var_11437_equation_0 = const()[name = tensor<string, []>("op_11437_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11437_cast = einsum(equation = var_11437_equation_0, values = (var_11323_cast, var_11281_cast))[name = tensor<string, []>("op_11437_cast")];
            tensor<fp16, []> var_11438_to_fp16 = const()[name = tensor<string, []>("op_11438_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_597_cast = mul(x = var_11437_cast, y = var_11438_to_fp16)[name = tensor<string, []>("aw_chunk_597_cast")];
            tensor<string, []> var_11441_equation_0 = const()[name = tensor<string, []>("op_11441_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_11441_cast = einsum(equation = var_11441_equation_0, values = (var_11323_cast, var_11282_cast))[name = tensor<string, []>("op_11441_cast")];
            tensor<fp16, []> var_11442_to_fp16 = const()[name = tensor<string, []>("op_11442_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_599_cast = mul(x = var_11441_cast, y = var_11442_to_fp16)[name = tensor<string, []>("aw_chunk_599_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11444_cast = softmax(axis = var_10257, x = aw_chunk_561_cast)[name = tensor<string, []>("op_11444_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11445_cast = softmax(axis = var_10257, x = aw_chunk_563_cast)[name = tensor<string, []>("op_11445_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11446_cast = softmax(axis = var_10257, x = aw_chunk_565_cast)[name = tensor<string, []>("op_11446_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11447_cast = softmax(axis = var_10257, x = aw_chunk_567_cast)[name = tensor<string, []>("op_11447_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11448_cast = softmax(axis = var_10257, x = aw_chunk_569_cast)[name = tensor<string, []>("op_11448_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11449_cast = softmax(axis = var_10257, x = aw_chunk_571_cast)[name = tensor<string, []>("op_11449_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11450_cast = softmax(axis = var_10257, x = aw_chunk_573_cast)[name = tensor<string, []>("op_11450_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11451_cast = softmax(axis = var_10257, x = aw_chunk_575_cast)[name = tensor<string, []>("op_11451_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11452_cast = softmax(axis = var_10257, x = aw_chunk_577_cast)[name = tensor<string, []>("op_11452_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11453_cast = softmax(axis = var_10257, x = aw_chunk_579_cast)[name = tensor<string, []>("op_11453_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11454_cast = softmax(axis = var_10257, x = aw_chunk_581_cast)[name = tensor<string, []>("op_11454_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11455_cast = softmax(axis = var_10257, x = aw_chunk_583_cast)[name = tensor<string, []>("op_11455_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11456_cast = softmax(axis = var_10257, x = aw_chunk_585_cast)[name = tensor<string, []>("op_11456_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11457_cast = softmax(axis = var_10257, x = aw_chunk_587_cast)[name = tensor<string, []>("op_11457_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11458_cast = softmax(axis = var_10257, x = aw_chunk_589_cast)[name = tensor<string, []>("op_11458_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11459_cast = softmax(axis = var_10257, x = aw_chunk_591_cast)[name = tensor<string, []>("op_11459_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11460_cast = softmax(axis = var_10257, x = aw_chunk_593_cast)[name = tensor<string, []>("op_11460_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11461_cast = softmax(axis = var_10257, x = aw_chunk_595_cast)[name = tensor<string, []>("op_11461_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11462_cast = softmax(axis = var_10257, x = aw_chunk_597_cast)[name = tensor<string, []>("op_11462_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_11463_cast = softmax(axis = var_10257, x = aw_chunk_599_cast)[name = tensor<string, []>("op_11463_cast")];
            tensor<string, []> var_11465_equation_0 = const()[name = tensor<string, []>("op_11465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11465_cast = einsum(equation = var_11465_equation_0, values = (var_11325_cast, var_11444_cast))[name = tensor<string, []>("op_11465_cast")];
            tensor<string, []> var_11467_equation_0 = const()[name = tensor<string, []>("op_11467_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11467_cast = einsum(equation = var_11467_equation_0, values = (var_11325_cast, var_11445_cast))[name = tensor<string, []>("op_11467_cast")];
            tensor<string, []> var_11469_equation_0 = const()[name = tensor<string, []>("op_11469_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11469_cast = einsum(equation = var_11469_equation_0, values = (var_11329_cast, var_11446_cast))[name = tensor<string, []>("op_11469_cast")];
            tensor<string, []> var_11471_equation_0 = const()[name = tensor<string, []>("op_11471_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11471_cast = einsum(equation = var_11471_equation_0, values = (var_11329_cast, var_11447_cast))[name = tensor<string, []>("op_11471_cast")];
            tensor<string, []> var_11473_equation_0 = const()[name = tensor<string, []>("op_11473_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11473_cast = einsum(equation = var_11473_equation_0, values = (var_11333_cast, var_11448_cast))[name = tensor<string, []>("op_11473_cast")];
            tensor<string, []> var_11475_equation_0 = const()[name = tensor<string, []>("op_11475_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11475_cast = einsum(equation = var_11475_equation_0, values = (var_11333_cast, var_11449_cast))[name = tensor<string, []>("op_11475_cast")];
            tensor<string, []> var_11477_equation_0 = const()[name = tensor<string, []>("op_11477_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11477_cast = einsum(equation = var_11477_equation_0, values = (var_11337_cast, var_11450_cast))[name = tensor<string, []>("op_11477_cast")];
            tensor<string, []> var_11479_equation_0 = const()[name = tensor<string, []>("op_11479_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11479_cast = einsum(equation = var_11479_equation_0, values = (var_11337_cast, var_11451_cast))[name = tensor<string, []>("op_11479_cast")];
            tensor<string, []> var_11481_equation_0 = const()[name = tensor<string, []>("op_11481_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11481_cast = einsum(equation = var_11481_equation_0, values = (var_11341_cast, var_11452_cast))[name = tensor<string, []>("op_11481_cast")];
            tensor<string, []> var_11483_equation_0 = const()[name = tensor<string, []>("op_11483_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11483_cast = einsum(equation = var_11483_equation_0, values = (var_11341_cast, var_11453_cast))[name = tensor<string, []>("op_11483_cast")];
            tensor<string, []> var_11485_equation_0 = const()[name = tensor<string, []>("op_11485_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11485_cast = einsum(equation = var_11485_equation_0, values = (var_11345_cast, var_11454_cast))[name = tensor<string, []>("op_11485_cast")];
            tensor<string, []> var_11487_equation_0 = const()[name = tensor<string, []>("op_11487_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11487_cast = einsum(equation = var_11487_equation_0, values = (var_11345_cast, var_11455_cast))[name = tensor<string, []>("op_11487_cast")];
            tensor<string, []> var_11489_equation_0 = const()[name = tensor<string, []>("op_11489_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11489_cast = einsum(equation = var_11489_equation_0, values = (var_11349_cast, var_11456_cast))[name = tensor<string, []>("op_11489_cast")];
            tensor<string, []> var_11491_equation_0 = const()[name = tensor<string, []>("op_11491_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11491_cast = einsum(equation = var_11491_equation_0, values = (var_11349_cast, var_11457_cast))[name = tensor<string, []>("op_11491_cast")];
            tensor<string, []> var_11493_equation_0 = const()[name = tensor<string, []>("op_11493_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11493_cast = einsum(equation = var_11493_equation_0, values = (var_11353_cast, var_11458_cast))[name = tensor<string, []>("op_11493_cast")];
            tensor<string, []> var_11495_equation_0 = const()[name = tensor<string, []>("op_11495_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11495_cast = einsum(equation = var_11495_equation_0, values = (var_11353_cast, var_11459_cast))[name = tensor<string, []>("op_11495_cast")];
            tensor<string, []> var_11497_equation_0 = const()[name = tensor<string, []>("op_11497_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11497_cast = einsum(equation = var_11497_equation_0, values = (var_11357_cast, var_11460_cast))[name = tensor<string, []>("op_11497_cast")];
            tensor<string, []> var_11499_equation_0 = const()[name = tensor<string, []>("op_11499_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11499_cast = einsum(equation = var_11499_equation_0, values = (var_11357_cast, var_11461_cast))[name = tensor<string, []>("op_11499_cast")];
            tensor<string, []> var_11501_equation_0 = const()[name = tensor<string, []>("op_11501_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11501_cast = einsum(equation = var_11501_equation_0, values = (var_11361_cast, var_11462_cast))[name = tensor<string, []>("op_11501_cast")];
            tensor<string, []> var_11503_equation_0 = const()[name = tensor<string, []>("op_11503_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11503_cast = einsum(equation = var_11503_equation_0, values = (var_11361_cast, var_11463_cast))[name = tensor<string, []>("op_11503_cast")];
            tensor<bool, []> var_11505_interleave_0 = const()[name = tensor<string, []>("op_11505_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11505_cast = concat(axis = var_10234, interleave = var_11505_interleave_0, values = (var_11465_cast, var_11467_cast))[name = tensor<string, []>("op_11505_cast")];
            tensor<bool, []> var_11507_interleave_0 = const()[name = tensor<string, []>("op_11507_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11507_cast = concat(axis = var_10234, interleave = var_11507_interleave_0, values = (var_11469_cast, var_11471_cast))[name = tensor<string, []>("op_11507_cast")];
            tensor<bool, []> var_11509_interleave_0 = const()[name = tensor<string, []>("op_11509_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11509_cast = concat(axis = var_10234, interleave = var_11509_interleave_0, values = (var_11473_cast, var_11475_cast))[name = tensor<string, []>("op_11509_cast")];
            tensor<bool, []> var_11511_interleave_0 = const()[name = tensor<string, []>("op_11511_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11511_cast = concat(axis = var_10234, interleave = var_11511_interleave_0, values = (var_11477_cast, var_11479_cast))[name = tensor<string, []>("op_11511_cast")];
            tensor<bool, []> var_11513_interleave_0 = const()[name = tensor<string, []>("op_11513_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11513_cast = concat(axis = var_10234, interleave = var_11513_interleave_0, values = (var_11481_cast, var_11483_cast))[name = tensor<string, []>("op_11513_cast")];
            tensor<bool, []> var_11515_interleave_0 = const()[name = tensor<string, []>("op_11515_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11515_cast = concat(axis = var_10234, interleave = var_11515_interleave_0, values = (var_11485_cast, var_11487_cast))[name = tensor<string, []>("op_11515_cast")];
            tensor<bool, []> var_11517_interleave_0 = const()[name = tensor<string, []>("op_11517_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11517_cast = concat(axis = var_10234, interleave = var_11517_interleave_0, values = (var_11489_cast, var_11491_cast))[name = tensor<string, []>("op_11517_cast")];
            tensor<bool, []> var_11519_interleave_0 = const()[name = tensor<string, []>("op_11519_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11519_cast = concat(axis = var_10234, interleave = var_11519_interleave_0, values = (var_11493_cast, var_11495_cast))[name = tensor<string, []>("op_11519_cast")];
            tensor<bool, []> var_11521_interleave_0 = const()[name = tensor<string, []>("op_11521_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11521_cast = concat(axis = var_10234, interleave = var_11521_interleave_0, values = (var_11497_cast, var_11499_cast))[name = tensor<string, []>("op_11521_cast")];
            tensor<bool, []> var_11523_interleave_0 = const()[name = tensor<string, []>("op_11523_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11523_cast = concat(axis = var_10234, interleave = var_11523_interleave_0, values = (var_11501_cast, var_11503_cast))[name = tensor<string, []>("op_11523_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_10257, interleave = input_409_interleave_0, values = (var_11505_cast, var_11507_cast, var_11509_cast, var_11511_cast, var_11513_cast, var_11515_cast, var_11517_cast, var_11519_cast, var_11521_cast, var_11523_cast))[name = tensor<string, []>("input_409_cast")];
            tensor<int32, [2]> var_11529 = const()[name = tensor<string, []>("op_11529"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11531 = const()[name = tensor<string, []>("op_11531"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11533_pad_type_0 = const()[name = tensor<string, []>("op_11533_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11533_pad_0 = const()[name = tensor<string, []>("op_11533_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204249152))), lut = tensor<fp16, [4]>([-0x1.844p-4, -0x1.c7cp-6, 0x1.d54p-6, 0x1.88p-4]), 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, []>(204351616)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11533_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_11531, groups = var_10257, pad = var_11533_pad_0, pad_type = var_11533_pad_type_0, strides = var_11529, 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_11533_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_69_cast = add(x = var_11533_cast, y = inputs_67_cast)[name = tensor<string, []>("inputs_69_cast")];
            tensor<int32, [1]> var_11537 = const()[name = tensor<string, []>("op_11537"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_69_cast = reduce_mean(axes = var_11537, keep_dims = var_10252, 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_11541 = const()[name = tensor<string, []>("op_11541"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11542_cast = reduce_mean(axes = var_11541, keep_dims = var_10252, x = zero_mean_sq_69_cast)[name = tensor<string, []>("op_11542_cast")];
            tensor<fp16, []> var_11543_to_fp16 = const()[name = tensor<string, []>("op_11543_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11544_cast = add(x = var_11542_cast, y = var_11543_to_fp16)[name = tensor<string, []>("op_11544_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_11544_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_11548_to_fp16 = const()[name = tensor<string, []>("op_11548_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204352960)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11549_cast = add(x = out_69_cast, y = var_11548_to_fp16)[name = tensor<string, []>("op_11549_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_11551_to_fp16 = const()[name = tensor<string, []>("op_11551_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204354304)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_249_cast = mul(x = var_11549_cast, y = var_11551_to_fp16)[name = tensor<string, []>("hidden_states_249_cast")];
            tensor<int32, [2]> var_11558 = const()[name = tensor<string, []>("op_11558"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11560 = const()[name = tensor<string, []>("op_11560"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204355648))), lut = tensor<fp16, [4]>([-0x1.794p-4, -0x1.c38p-6, 0x1.bbcp-6, 0x1.768p-4]), 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_11560, groups = var_10257, pad = q_47_pad_0, pad_type = q_47_pad_type_0, strides = var_11558, 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_11564 = const()[name = tensor<string, []>("op_11564"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11566 = const()[name = tensor<string, []>("op_11566"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_93_pad_type_0 = const()[name = tensor<string, []>("k_93_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_93_pad_0 = const()[name = tensor<string, []>("k_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204458112))), lut = tensor<fp16, [4]>([-0x1.52p-4, -0x1.8c8p-6, 0x1.978p-6, 0x1.54p-4]), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_93_cast = conv(dilations = var_11566, groups = var_10257, pad = k_93_pad_0, pad_type = k_93_pad_type_0, strides = var_11564, 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_11570 = const()[name = tensor<string, []>("op_11570"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11572 = const()[name = tensor<string, []>("op_11572"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_47_pad_type_0 = const()[name = tensor<string, []>("v_47_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_47_pad_0 = const()[name = tensor<string, []>("v_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204622016))), lut = tensor<fp16, [4]>([-0x1.c2cp-5, -0x1.014p-6, 0x1.044p-6, 0x1.c48p-5]), name = tensor<string, []>("up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_47_cast = conv(dilations = var_11572, groups = var_10257, pad = v_47_pad_0, pad_type = v_47_pad_type_0, strides = var_11570, 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_11576_begin_0 = const()[name = tensor<string, []>("op_11576_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11576_end_0 = const()[name = tensor<string, []>("op_11576_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11576_end_mask_0 = const()[name = tensor<string, []>("op_11576_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11576_cast = slice_by_index(begin = var_11576_begin_0, end = var_11576_end_0, end_mask = var_11576_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11576_cast")];
            tensor<int32, [4]> var_11580_begin_0 = const()[name = tensor<string, []>("op_11580_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11580_end_0 = const()[name = tensor<string, []>("op_11580_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_11580_end_mask_0 = const()[name = tensor<string, []>("op_11580_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11580_cast = slice_by_index(begin = var_11580_begin_0, end = var_11580_end_0, end_mask = var_11580_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11580_cast")];
            tensor<int32, [4]> var_11584_begin_0 = const()[name = tensor<string, []>("op_11584_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11584_end_0 = const()[name = tensor<string, []>("op_11584_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_11584_end_mask_0 = const()[name = tensor<string, []>("op_11584_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11584_cast = slice_by_index(begin = var_11584_begin_0, end = var_11584_end_0, end_mask = var_11584_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11584_cast")];
            tensor<int32, [4]> var_11588_begin_0 = const()[name = tensor<string, []>("op_11588_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11588_end_0 = const()[name = tensor<string, []>("op_11588_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_11588_end_mask_0 = const()[name = tensor<string, []>("op_11588_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11588_cast = slice_by_index(begin = var_11588_begin_0, end = var_11588_end_0, end_mask = var_11588_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11588_cast")];
            tensor<int32, [4]> var_11592_begin_0 = const()[name = tensor<string, []>("op_11592_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11592_end_0 = const()[name = tensor<string, []>("op_11592_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_11592_end_mask_0 = const()[name = tensor<string, []>("op_11592_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11592_cast = slice_by_index(begin = var_11592_begin_0, end = var_11592_end_0, end_mask = var_11592_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11592_cast")];
            tensor<int32, [4]> var_11596_begin_0 = const()[name = tensor<string, []>("op_11596_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11596_end_0 = const()[name = tensor<string, []>("op_11596_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_11596_end_mask_0 = const()[name = tensor<string, []>("op_11596_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11596_cast = slice_by_index(begin = var_11596_begin_0, end = var_11596_end_0, end_mask = var_11596_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11596_cast")];
            tensor<int32, [4]> var_11600_begin_0 = const()[name = tensor<string, []>("op_11600_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11600_end_0 = const()[name = tensor<string, []>("op_11600_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_11600_end_mask_0 = const()[name = tensor<string, []>("op_11600_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11600_cast = slice_by_index(begin = var_11600_begin_0, end = var_11600_end_0, end_mask = var_11600_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11600_cast")];
            tensor<int32, [4]> var_11604_begin_0 = const()[name = tensor<string, []>("op_11604_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11604_end_0 = const()[name = tensor<string, []>("op_11604_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_11604_end_mask_0 = const()[name = tensor<string, []>("op_11604_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11604_cast = slice_by_index(begin = var_11604_begin_0, end = var_11604_end_0, end_mask = var_11604_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11604_cast")];
            tensor<int32, [4]> var_11608_begin_0 = const()[name = tensor<string, []>("op_11608_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11608_end_0 = const()[name = tensor<string, []>("op_11608_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_11608_end_mask_0 = const()[name = tensor<string, []>("op_11608_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11608_cast = slice_by_index(begin = var_11608_begin_0, end = var_11608_end_0, end_mask = var_11608_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11608_cast")];
            tensor<int32, [4]> var_11612_begin_0 = const()[name = tensor<string, []>("op_11612_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11612_end_0 = const()[name = tensor<string, []>("op_11612_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_11612_end_mask_0 = const()[name = tensor<string, []>("op_11612_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_11612_cast = slice_by_index(begin = var_11612_begin_0, end = var_11612_end_0, end_mask = var_11612_end_mask_0, x = q_47_cast)[name = tensor<string, []>("op_11612_cast")];
            tensor<int32, [4]> var_11615_begin_0 = const()[name = tensor<string, []>("op_11615_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11615_end_0 = const()[name = tensor<string, []>("op_11615_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11615_end_mask_0 = const()[name = tensor<string, []>("op_11615_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11615_cast = slice_by_index(begin = var_11615_begin_0, end = var_11615_end_0, end_mask = var_11615_end_mask_0, x = var_11576_cast)[name = tensor<string, []>("op_11615_cast")];
            tensor<int32, [4]> var_11616_begin_0 = const()[name = tensor<string, []>("op_11616_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11616_end_0 = const()[name = tensor<string, []>("op_11616_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11616_end_mask_0 = const()[name = tensor<string, []>("op_11616_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11616_cast = slice_by_index(begin = var_11616_begin_0, end = var_11616_end_0, end_mask = var_11616_end_mask_0, x = var_11576_cast)[name = tensor<string, []>("op_11616_cast")];
            tensor<int32, [4]> var_11617_begin_0 = const()[name = tensor<string, []>("op_11617_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11617_end_0 = const()[name = tensor<string, []>("op_11617_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11617_end_mask_0 = const()[name = tensor<string, []>("op_11617_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11617_cast = slice_by_index(begin = var_11617_begin_0, end = var_11617_end_0, end_mask = var_11617_end_mask_0, x = var_11580_cast)[name = tensor<string, []>("op_11617_cast")];
            tensor<int32, [4]> var_11618_begin_0 = const()[name = tensor<string, []>("op_11618_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11618_end_0 = const()[name = tensor<string, []>("op_11618_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11618_end_mask_0 = const()[name = tensor<string, []>("op_11618_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11618_cast = slice_by_index(begin = var_11618_begin_0, end = var_11618_end_0, end_mask = var_11618_end_mask_0, x = var_11580_cast)[name = tensor<string, []>("op_11618_cast")];
            tensor<int32, [4]> var_11619_begin_0 = const()[name = tensor<string, []>("op_11619_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11619_end_0 = const()[name = tensor<string, []>("op_11619_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11619_end_mask_0 = const()[name = tensor<string, []>("op_11619_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11619_cast = slice_by_index(begin = var_11619_begin_0, end = var_11619_end_0, end_mask = var_11619_end_mask_0, x = var_11584_cast)[name = tensor<string, []>("op_11619_cast")];
            tensor<int32, [4]> var_11620_begin_0 = const()[name = tensor<string, []>("op_11620_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11620_end_0 = const()[name = tensor<string, []>("op_11620_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11620_end_mask_0 = const()[name = tensor<string, []>("op_11620_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11620_cast = slice_by_index(begin = var_11620_begin_0, end = var_11620_end_0, end_mask = var_11620_end_mask_0, x = var_11584_cast)[name = tensor<string, []>("op_11620_cast")];
            tensor<int32, [4]> var_11621_begin_0 = const()[name = tensor<string, []>("op_11621_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11621_end_0 = const()[name = tensor<string, []>("op_11621_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11621_end_mask_0 = const()[name = tensor<string, []>("op_11621_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11621_cast = slice_by_index(begin = var_11621_begin_0, end = var_11621_end_0, end_mask = var_11621_end_mask_0, x = var_11588_cast)[name = tensor<string, []>("op_11621_cast")];
            tensor<int32, [4]> var_11622_begin_0 = const()[name = tensor<string, []>("op_11622_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11622_end_0 = const()[name = tensor<string, []>("op_11622_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11622_end_mask_0 = const()[name = tensor<string, []>("op_11622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11622_cast = slice_by_index(begin = var_11622_begin_0, end = var_11622_end_0, end_mask = var_11622_end_mask_0, x = var_11588_cast)[name = tensor<string, []>("op_11622_cast")];
            tensor<int32, [4]> var_11623_begin_0 = const()[name = tensor<string, []>("op_11623_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11623_end_0 = const()[name = tensor<string, []>("op_11623_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11623_end_mask_0 = const()[name = tensor<string, []>("op_11623_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11623_cast = slice_by_index(begin = var_11623_begin_0, end = var_11623_end_0, end_mask = var_11623_end_mask_0, x = var_11592_cast)[name = tensor<string, []>("op_11623_cast")];
            tensor<int32, [4]> var_11624_begin_0 = const()[name = tensor<string, []>("op_11624_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11624_end_0 = const()[name = tensor<string, []>("op_11624_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11624_end_mask_0 = const()[name = tensor<string, []>("op_11624_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11624_cast = slice_by_index(begin = var_11624_begin_0, end = var_11624_end_0, end_mask = var_11624_end_mask_0, x = var_11592_cast)[name = tensor<string, []>("op_11624_cast")];
            tensor<int32, [4]> var_11625_begin_0 = const()[name = tensor<string, []>("op_11625_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11625_end_0 = const()[name = tensor<string, []>("op_11625_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11625_end_mask_0 = const()[name = tensor<string, []>("op_11625_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11625_cast = slice_by_index(begin = var_11625_begin_0, end = var_11625_end_0, end_mask = var_11625_end_mask_0, x = var_11596_cast)[name = tensor<string, []>("op_11625_cast")];
            tensor<int32, [4]> var_11626_begin_0 = const()[name = tensor<string, []>("op_11626_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11626_end_0 = const()[name = tensor<string, []>("op_11626_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11626_end_mask_0 = const()[name = tensor<string, []>("op_11626_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11626_cast = slice_by_index(begin = var_11626_begin_0, end = var_11626_end_0, end_mask = var_11626_end_mask_0, x = var_11596_cast)[name = tensor<string, []>("op_11626_cast")];
            tensor<int32, [4]> var_11627_begin_0 = const()[name = tensor<string, []>("op_11627_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11627_end_0 = const()[name = tensor<string, []>("op_11627_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11627_end_mask_0 = const()[name = tensor<string, []>("op_11627_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11627_cast = slice_by_index(begin = var_11627_begin_0, end = var_11627_end_0, end_mask = var_11627_end_mask_0, x = var_11600_cast)[name = tensor<string, []>("op_11627_cast")];
            tensor<int32, [4]> var_11628_begin_0 = const()[name = tensor<string, []>("op_11628_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11628_end_0 = const()[name = tensor<string, []>("op_11628_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11628_end_mask_0 = const()[name = tensor<string, []>("op_11628_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11628_cast = slice_by_index(begin = var_11628_begin_0, end = var_11628_end_0, end_mask = var_11628_end_mask_0, x = var_11600_cast)[name = tensor<string, []>("op_11628_cast")];
            tensor<int32, [4]> var_11629_begin_0 = const()[name = tensor<string, []>("op_11629_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11629_end_0 = const()[name = tensor<string, []>("op_11629_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11629_end_mask_0 = const()[name = tensor<string, []>("op_11629_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11629_cast = slice_by_index(begin = var_11629_begin_0, end = var_11629_end_0, end_mask = var_11629_end_mask_0, x = var_11604_cast)[name = tensor<string, []>("op_11629_cast")];
            tensor<int32, [4]> var_11630_begin_0 = const()[name = tensor<string, []>("op_11630_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11630_end_0 = const()[name = tensor<string, []>("op_11630_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11630_end_mask_0 = const()[name = tensor<string, []>("op_11630_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11630_cast = slice_by_index(begin = var_11630_begin_0, end = var_11630_end_0, end_mask = var_11630_end_mask_0, x = var_11604_cast)[name = tensor<string, []>("op_11630_cast")];
            tensor<int32, [4]> var_11631_begin_0 = const()[name = tensor<string, []>("op_11631_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11631_end_0 = const()[name = tensor<string, []>("op_11631_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11631_end_mask_0 = const()[name = tensor<string, []>("op_11631_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11631_cast = slice_by_index(begin = var_11631_begin_0, end = var_11631_end_0, end_mask = var_11631_end_mask_0, x = var_11608_cast)[name = tensor<string, []>("op_11631_cast")];
            tensor<int32, [4]> var_11632_begin_0 = const()[name = tensor<string, []>("op_11632_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11632_end_0 = const()[name = tensor<string, []>("op_11632_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11632_end_mask_0 = const()[name = tensor<string, []>("op_11632_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11632_cast = slice_by_index(begin = var_11632_begin_0, end = var_11632_end_0, end_mask = var_11632_end_mask_0, x = var_11608_cast)[name = tensor<string, []>("op_11632_cast")];
            tensor<int32, [4]> var_11633_begin_0 = const()[name = tensor<string, []>("op_11633_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11633_end_0 = const()[name = tensor<string, []>("op_11633_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_11633_end_mask_0 = const()[name = tensor<string, []>("op_11633_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11633_cast = slice_by_index(begin = var_11633_begin_0, end = var_11633_end_0, end_mask = var_11633_end_mask_0, x = var_11612_cast)[name = tensor<string, []>("op_11633_cast")];
            tensor<int32, [4]> var_11634_begin_0 = const()[name = tensor<string, []>("op_11634_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11634_end_0 = const()[name = tensor<string, []>("op_11634_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_11634_end_mask_0 = const()[name = tensor<string, []>("op_11634_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_11634_cast = slice_by_index(begin = var_11634_begin_0, end = var_11634_end_0, end_mask = var_11634_end_mask_0, x = var_11612_cast)[name = tensor<string, []>("op_11634_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_11639_begin_0 = const()[name = tensor<string, []>("op_11639_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11639_end_0 = const()[name = tensor<string, []>("op_11639_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_11639_end_mask_0 = const()[name = tensor<string, []>("op_11639_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, 64]> var_11639_cast = slice_by_index(begin = var_11639_begin_0, end = var_11639_end_0, end_mask = var_11639_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11639_cast")];
            tensor<int32, [4]> var_11643_begin_0 = const()[name = tensor<string, []>("op_11643_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_11643_end_0 = const()[name = tensor<string, []>("op_11643_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_11643_end_mask_0 = const()[name = tensor<string, []>("op_11643_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11643_cast = slice_by_index(begin = var_11643_begin_0, end = var_11643_end_0, end_mask = var_11643_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11643_cast")];
            tensor<int32, [4]> var_11647_begin_0 = const()[name = tensor<string, []>("op_11647_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_11647_end_0 = const()[name = tensor<string, []>("op_11647_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_11647_end_mask_0 = const()[name = tensor<string, []>("op_11647_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11647_cast = slice_by_index(begin = var_11647_begin_0, end = var_11647_end_0, end_mask = var_11647_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11647_cast")];
            tensor<int32, [4]> var_11651_begin_0 = const()[name = tensor<string, []>("op_11651_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_11651_end_0 = const()[name = tensor<string, []>("op_11651_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_11651_end_mask_0 = const()[name = tensor<string, []>("op_11651_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11651_cast = slice_by_index(begin = var_11651_begin_0, end = var_11651_end_0, end_mask = var_11651_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11651_cast")];
            tensor<int32, [4]> var_11655_begin_0 = const()[name = tensor<string, []>("op_11655_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_11655_end_0 = const()[name = tensor<string, []>("op_11655_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_11655_end_mask_0 = const()[name = tensor<string, []>("op_11655_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11655_cast = slice_by_index(begin = var_11655_begin_0, end = var_11655_end_0, end_mask = var_11655_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11655_cast")];
            tensor<int32, [4]> var_11659_begin_0 = const()[name = tensor<string, []>("op_11659_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_11659_end_0 = const()[name = tensor<string, []>("op_11659_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_11659_end_mask_0 = const()[name = tensor<string, []>("op_11659_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11659_cast = slice_by_index(begin = var_11659_begin_0, end = var_11659_end_0, end_mask = var_11659_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11659_cast")];
            tensor<int32, [4]> var_11663_begin_0 = const()[name = tensor<string, []>("op_11663_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_11663_end_0 = const()[name = tensor<string, []>("op_11663_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_11663_end_mask_0 = const()[name = tensor<string, []>("op_11663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11663_cast = slice_by_index(begin = var_11663_begin_0, end = var_11663_end_0, end_mask = var_11663_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11663_cast")];
            tensor<int32, [4]> var_11667_begin_0 = const()[name = tensor<string, []>("op_11667_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_11667_end_0 = const()[name = tensor<string, []>("op_11667_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_11667_end_mask_0 = const()[name = tensor<string, []>("op_11667_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11667_cast = slice_by_index(begin = var_11667_begin_0, end = var_11667_end_0, end_mask = var_11667_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11667_cast")];
            tensor<int32, [4]> var_11671_begin_0 = const()[name = tensor<string, []>("op_11671_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_11671_end_0 = const()[name = tensor<string, []>("op_11671_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_11671_end_mask_0 = const()[name = tensor<string, []>("op_11671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11671_cast = slice_by_index(begin = var_11671_begin_0, end = var_11671_end_0, end_mask = var_11671_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11671_cast")];
            tensor<int32, [4]> var_11675_begin_0 = const()[name = tensor<string, []>("op_11675_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_11675_end_0 = const()[name = tensor<string, []>("op_11675_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_11675_end_mask_0 = const()[name = tensor<string, []>("op_11675_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_11675_cast = slice_by_index(begin = var_11675_begin_0, end = var_11675_end_0, end_mask = var_11675_end_mask_0, x = transpose_8)[name = tensor<string, []>("op_11675_cast")];
            tensor<int32, [4]> var_11677_begin_0 = const()[name = tensor<string, []>("op_11677_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_11677_end_0 = const()[name = tensor<string, []>("op_11677_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_11677_end_mask_0 = const()[name = tensor<string, []>("op_11677_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11677_cast = slice_by_index(begin = var_11677_begin_0, end = var_11677_end_0, end_mask = var_11677_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11677_cast")];
            tensor<int32, [4]> var_11681_begin_0 = const()[name = tensor<string, []>("op_11681_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_11681_end_0 = const()[name = tensor<string, []>("op_11681_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_11681_end_mask_0 = const()[name = tensor<string, []>("op_11681_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11681_cast = slice_by_index(begin = var_11681_begin_0, end = var_11681_end_0, end_mask = var_11681_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11681_cast")];
            tensor<int32, [4]> var_11685_begin_0 = const()[name = tensor<string, []>("op_11685_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_11685_end_0 = const()[name = tensor<string, []>("op_11685_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_11685_end_mask_0 = const()[name = tensor<string, []>("op_11685_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11685_cast = slice_by_index(begin = var_11685_begin_0, end = var_11685_end_0, end_mask = var_11685_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11685_cast")];
            tensor<int32, [4]> var_11689_begin_0 = const()[name = tensor<string, []>("op_11689_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_11689_end_0 = const()[name = tensor<string, []>("op_11689_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_11689_end_mask_0 = const()[name = tensor<string, []>("op_11689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11689_cast = slice_by_index(begin = var_11689_begin_0, end = var_11689_end_0, end_mask = var_11689_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11689_cast")];
            tensor<int32, [4]> var_11693_begin_0 = const()[name = tensor<string, []>("op_11693_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_11693_end_0 = const()[name = tensor<string, []>("op_11693_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_11693_end_mask_0 = const()[name = tensor<string, []>("op_11693_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11693_cast = slice_by_index(begin = var_11693_begin_0, end = var_11693_end_0, end_mask = var_11693_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11693_cast")];
            tensor<int32, [4]> var_11697_begin_0 = const()[name = tensor<string, []>("op_11697_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_11697_end_0 = const()[name = tensor<string, []>("op_11697_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_11697_end_mask_0 = const()[name = tensor<string, []>("op_11697_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11697_cast = slice_by_index(begin = var_11697_begin_0, end = var_11697_end_0, end_mask = var_11697_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11697_cast")];
            tensor<int32, [4]> var_11701_begin_0 = const()[name = tensor<string, []>("op_11701_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_11701_end_0 = const()[name = tensor<string, []>("op_11701_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_11701_end_mask_0 = const()[name = tensor<string, []>("op_11701_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11701_cast = slice_by_index(begin = var_11701_begin_0, end = var_11701_end_0, end_mask = var_11701_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11701_cast")];
            tensor<int32, [4]> var_11705_begin_0 = const()[name = tensor<string, []>("op_11705_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_11705_end_0 = const()[name = tensor<string, []>("op_11705_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_11705_end_mask_0 = const()[name = tensor<string, []>("op_11705_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11705_cast = slice_by_index(begin = var_11705_begin_0, end = var_11705_end_0, end_mask = var_11705_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11705_cast")];
            tensor<int32, [4]> var_11709_begin_0 = const()[name = tensor<string, []>("op_11709_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_11709_end_0 = const()[name = tensor<string, []>("op_11709_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_11709_end_mask_0 = const()[name = tensor<string, []>("op_11709_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11709_cast = slice_by_index(begin = var_11709_begin_0, end = var_11709_end_0, end_mask = var_11709_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11709_cast")];
            tensor<int32, [4]> var_11713_begin_0 = const()[name = tensor<string, []>("op_11713_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_11713_end_0 = const()[name = tensor<string, []>("op_11713_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_11713_end_mask_0 = const()[name = tensor<string, []>("op_11713_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_11713_cast = slice_by_index(begin = var_11713_begin_0, end = var_11713_end_0, end_mask = var_11713_end_mask_0, x = v_47_cast)[name = tensor<string, []>("op_11713_cast")];
            tensor<string, []> var_11717_equation_0 = const()[name = tensor<string, []>("op_11717_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11717_cast = einsum(equation = var_11717_equation_0, values = (var_11639_cast, var_11615_cast))[name = tensor<string, []>("op_11717_cast")];
            tensor<fp16, []> var_11718_to_fp16 = const()[name = tensor<string, []>("op_11718_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_601_cast = mul(x = var_11717_cast, y = var_11718_to_fp16)[name = tensor<string, []>("aw_chunk_601_cast")];
            tensor<string, []> var_11721_equation_0 = const()[name = tensor<string, []>("op_11721_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11721_cast = einsum(equation = var_11721_equation_0, values = (var_11639_cast, var_11616_cast))[name = tensor<string, []>("op_11721_cast")];
            tensor<fp16, []> var_11722_to_fp16 = const()[name = tensor<string, []>("op_11722_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_603_cast = mul(x = var_11721_cast, y = var_11722_to_fp16)[name = tensor<string, []>("aw_chunk_603_cast")];
            tensor<string, []> var_11725_equation_0 = const()[name = tensor<string, []>("op_11725_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11725_cast = einsum(equation = var_11725_equation_0, values = (var_11643_cast, var_11617_cast))[name = tensor<string, []>("op_11725_cast")];
            tensor<fp16, []> var_11726_to_fp16 = const()[name = tensor<string, []>("op_11726_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_605_cast = mul(x = var_11725_cast, y = var_11726_to_fp16)[name = tensor<string, []>("aw_chunk_605_cast")];
            tensor<string, []> var_11729_equation_0 = const()[name = tensor<string, []>("op_11729_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11729_cast = einsum(equation = var_11729_equation_0, values = (var_11643_cast, var_11618_cast))[name = tensor<string, []>("op_11729_cast")];
            tensor<fp16, []> var_11730_to_fp16 = const()[name = tensor<string, []>("op_11730_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_607_cast = mul(x = var_11729_cast, y = var_11730_to_fp16)[name = tensor<string, []>("aw_chunk_607_cast")];
            tensor<string, []> var_11733_equation_0 = const()[name = tensor<string, []>("op_11733_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11733_cast = einsum(equation = var_11733_equation_0, values = (var_11647_cast, var_11619_cast))[name = tensor<string, []>("op_11733_cast")];
            tensor<fp16, []> var_11734_to_fp16 = const()[name = tensor<string, []>("op_11734_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_609_cast = mul(x = var_11733_cast, y = var_11734_to_fp16)[name = tensor<string, []>("aw_chunk_609_cast")];
            tensor<string, []> var_11737_equation_0 = const()[name = tensor<string, []>("op_11737_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11737_cast = einsum(equation = var_11737_equation_0, values = (var_11647_cast, var_11620_cast))[name = tensor<string, []>("op_11737_cast")];
            tensor<fp16, []> var_11738_to_fp16 = const()[name = tensor<string, []>("op_11738_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_611_cast = mul(x = var_11737_cast, y = var_11738_to_fp16)[name = tensor<string, []>("aw_chunk_611_cast")];
            tensor<string, []> var_11741_equation_0 = const()[name = tensor<string, []>("op_11741_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11741_cast = einsum(equation = var_11741_equation_0, values = (var_11651_cast, var_11621_cast))[name = tensor<string, []>("op_11741_cast")];
            tensor<fp16, []> var_11742_to_fp16 = const()[name = tensor<string, []>("op_11742_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_613_cast = mul(x = var_11741_cast, y = var_11742_to_fp16)[name = tensor<string, []>("aw_chunk_613_cast")];
            tensor<string, []> var_11745_equation_0 = const()[name = tensor<string, []>("op_11745_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11745_cast = einsum(equation = var_11745_equation_0, values = (var_11651_cast, var_11622_cast))[name = tensor<string, []>("op_11745_cast")];
            tensor<fp16, []> var_11746_to_fp16 = const()[name = tensor<string, []>("op_11746_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_615_cast = mul(x = var_11745_cast, y = var_11746_to_fp16)[name = tensor<string, []>("aw_chunk_615_cast")];
            tensor<string, []> var_11749_equation_0 = const()[name = tensor<string, []>("op_11749_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11749_cast = einsum(equation = var_11749_equation_0, values = (var_11655_cast, var_11623_cast))[name = tensor<string, []>("op_11749_cast")];
            tensor<fp16, []> var_11750_to_fp16 = const()[name = tensor<string, []>("op_11750_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_617_cast = mul(x = var_11749_cast, y = var_11750_to_fp16)[name = tensor<string, []>("aw_chunk_617_cast")];
            tensor<string, []> var_11753_equation_0 = const()[name = tensor<string, []>("op_11753_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11753_cast = einsum(equation = var_11753_equation_0, values = (var_11655_cast, var_11624_cast))[name = tensor<string, []>("op_11753_cast")];
            tensor<fp16, []> var_11754_to_fp16 = const()[name = tensor<string, []>("op_11754_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_619_cast = mul(x = var_11753_cast, y = var_11754_to_fp16)[name = tensor<string, []>("aw_chunk_619_cast")];
            tensor<string, []> var_11757_equation_0 = const()[name = tensor<string, []>("op_11757_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11757_cast = einsum(equation = var_11757_equation_0, values = (var_11659_cast, var_11625_cast))[name = tensor<string, []>("op_11757_cast")];
            tensor<fp16, []> var_11758_to_fp16 = const()[name = tensor<string, []>("op_11758_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_621_cast = mul(x = var_11757_cast, y = var_11758_to_fp16)[name = tensor<string, []>("aw_chunk_621_cast")];
            tensor<string, []> var_11761_equation_0 = const()[name = tensor<string, []>("op_11761_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11761_cast = einsum(equation = var_11761_equation_0, values = (var_11659_cast, var_11626_cast))[name = tensor<string, []>("op_11761_cast")];
            tensor<fp16, []> var_11762_to_fp16 = const()[name = tensor<string, []>("op_11762_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_623_cast = mul(x = var_11761_cast, y = var_11762_to_fp16)[name = tensor<string, []>("aw_chunk_623_cast")];
            tensor<string, []> var_11765_equation_0 = const()[name = tensor<string, []>("op_11765_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11765_cast = einsum(equation = var_11765_equation_0, values = (var_11663_cast, var_11627_cast))[name = tensor<string, []>("op_11765_cast")];
            tensor<fp16, []> var_11766_to_fp16 = const()[name = tensor<string, []>("op_11766_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_625_cast = mul(x = var_11765_cast, y = var_11766_to_fp16)[name = tensor<string, []>("aw_chunk_625_cast")];
            tensor<string, []> var_11769_equation_0 = const()[name = tensor<string, []>("op_11769_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11769_cast = einsum(equation = var_11769_equation_0, values = (var_11663_cast, var_11628_cast))[name = tensor<string, []>("op_11769_cast")];
            tensor<fp16, []> var_11770_to_fp16 = const()[name = tensor<string, []>("op_11770_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_627_cast = mul(x = var_11769_cast, y = var_11770_to_fp16)[name = tensor<string, []>("aw_chunk_627_cast")];
            tensor<string, []> var_11773_equation_0 = const()[name = tensor<string, []>("op_11773_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11773_cast = einsum(equation = var_11773_equation_0, values = (var_11667_cast, var_11629_cast))[name = tensor<string, []>("op_11773_cast")];
            tensor<fp16, []> var_11774_to_fp16 = const()[name = tensor<string, []>("op_11774_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_629_cast = mul(x = var_11773_cast, y = var_11774_to_fp16)[name = tensor<string, []>("aw_chunk_629_cast")];
            tensor<string, []> var_11777_equation_0 = const()[name = tensor<string, []>("op_11777_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11777_cast = einsum(equation = var_11777_equation_0, values = (var_11667_cast, var_11630_cast))[name = tensor<string, []>("op_11777_cast")];
            tensor<fp16, []> var_11778_to_fp16 = const()[name = tensor<string, []>("op_11778_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_631_cast = mul(x = var_11777_cast, y = var_11778_to_fp16)[name = tensor<string, []>("aw_chunk_631_cast")];
            tensor<string, []> var_11781_equation_0 = const()[name = tensor<string, []>("op_11781_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11781_cast = einsum(equation = var_11781_equation_0, values = (var_11671_cast, var_11631_cast))[name = tensor<string, []>("op_11781_cast")];
            tensor<fp16, []> var_11782_to_fp16 = const()[name = tensor<string, []>("op_11782_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_633_cast = mul(x = var_11781_cast, y = var_11782_to_fp16)[name = tensor<string, []>("aw_chunk_633_cast")];
            tensor<string, []> var_11785_equation_0 = const()[name = tensor<string, []>("op_11785_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11785_cast = einsum(equation = var_11785_equation_0, values = (var_11671_cast, var_11632_cast))[name = tensor<string, []>("op_11785_cast")];
            tensor<fp16, []> var_11786_to_fp16 = const()[name = tensor<string, []>("op_11786_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_635_cast = mul(x = var_11785_cast, y = var_11786_to_fp16)[name = tensor<string, []>("aw_chunk_635_cast")];
            tensor<string, []> var_11789_equation_0 = const()[name = tensor<string, []>("op_11789_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11789_cast = einsum(equation = var_11789_equation_0, values = (var_11675_cast, var_11633_cast))[name = tensor<string, []>("op_11789_cast")];
            tensor<fp16, []> var_11790_to_fp16 = const()[name = tensor<string, []>("op_11790_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_637_cast = mul(x = var_11789_cast, y = var_11790_to_fp16)[name = tensor<string, []>("aw_chunk_637_cast")];
            tensor<string, []> var_11793_equation_0 = const()[name = tensor<string, []>("op_11793_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_11793_cast = einsum(equation = var_11793_equation_0, values = (var_11675_cast, var_11634_cast))[name = tensor<string, []>("op_11793_cast")];
            tensor<fp16, []> var_11794_to_fp16 = const()[name = tensor<string, []>("op_11794_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_639_cast = mul(x = var_11793_cast, y = var_11794_to_fp16)[name = tensor<string, []>("aw_chunk_639_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11796_cast = softmax(axis = var_10257, x = aw_chunk_601_cast)[name = tensor<string, []>("op_11796_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11797_cast = softmax(axis = var_10257, x = aw_chunk_603_cast)[name = tensor<string, []>("op_11797_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11798_cast = softmax(axis = var_10257, x = aw_chunk_605_cast)[name = tensor<string, []>("op_11798_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11799_cast = softmax(axis = var_10257, x = aw_chunk_607_cast)[name = tensor<string, []>("op_11799_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11800_cast = softmax(axis = var_10257, x = aw_chunk_609_cast)[name = tensor<string, []>("op_11800_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11801_cast = softmax(axis = var_10257, x = aw_chunk_611_cast)[name = tensor<string, []>("op_11801_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11802_cast = softmax(axis = var_10257, x = aw_chunk_613_cast)[name = tensor<string, []>("op_11802_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11803_cast = softmax(axis = var_10257, x = aw_chunk_615_cast)[name = tensor<string, []>("op_11803_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11804_cast = softmax(axis = var_10257, x = aw_chunk_617_cast)[name = tensor<string, []>("op_11804_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11805_cast = softmax(axis = var_10257, x = aw_chunk_619_cast)[name = tensor<string, []>("op_11805_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11806_cast = softmax(axis = var_10257, x = aw_chunk_621_cast)[name = tensor<string, []>("op_11806_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11807_cast = softmax(axis = var_10257, x = aw_chunk_623_cast)[name = tensor<string, []>("op_11807_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11808_cast = softmax(axis = var_10257, x = aw_chunk_625_cast)[name = tensor<string, []>("op_11808_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11809_cast = softmax(axis = var_10257, x = aw_chunk_627_cast)[name = tensor<string, []>("op_11809_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11810_cast = softmax(axis = var_10257, x = aw_chunk_629_cast)[name = tensor<string, []>("op_11810_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11811_cast = softmax(axis = var_10257, x = aw_chunk_631_cast)[name = tensor<string, []>("op_11811_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11812_cast = softmax(axis = var_10257, x = aw_chunk_633_cast)[name = tensor<string, []>("op_11812_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11813_cast = softmax(axis = var_10257, x = aw_chunk_635_cast)[name = tensor<string, []>("op_11813_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11814_cast = softmax(axis = var_10257, x = aw_chunk_637_cast)[name = tensor<string, []>("op_11814_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_11815_cast = softmax(axis = var_10257, x = aw_chunk_639_cast)[name = tensor<string, []>("op_11815_cast")];
            tensor<string, []> var_11817_equation_0 = const()[name = tensor<string, []>("op_11817_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11817_cast = einsum(equation = var_11817_equation_0, values = (var_11677_cast, var_11796_cast))[name = tensor<string, []>("op_11817_cast")];
            tensor<string, []> var_11819_equation_0 = const()[name = tensor<string, []>("op_11819_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11819_cast = einsum(equation = var_11819_equation_0, values = (var_11677_cast, var_11797_cast))[name = tensor<string, []>("op_11819_cast")];
            tensor<string, []> var_11821_equation_0 = const()[name = tensor<string, []>("op_11821_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11821_cast = einsum(equation = var_11821_equation_0, values = (var_11681_cast, var_11798_cast))[name = tensor<string, []>("op_11821_cast")];
            tensor<string, []> var_11823_equation_0 = const()[name = tensor<string, []>("op_11823_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11823_cast = einsum(equation = var_11823_equation_0, values = (var_11681_cast, var_11799_cast))[name = tensor<string, []>("op_11823_cast")];
            tensor<string, []> var_11825_equation_0 = const()[name = tensor<string, []>("op_11825_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11825_cast = einsum(equation = var_11825_equation_0, values = (var_11685_cast, var_11800_cast))[name = tensor<string, []>("op_11825_cast")];
            tensor<string, []> var_11827_equation_0 = const()[name = tensor<string, []>("op_11827_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11827_cast = einsum(equation = var_11827_equation_0, values = (var_11685_cast, var_11801_cast))[name = tensor<string, []>("op_11827_cast")];
            tensor<string, []> var_11829_equation_0 = const()[name = tensor<string, []>("op_11829_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11829_cast = einsum(equation = var_11829_equation_0, values = (var_11689_cast, var_11802_cast))[name = tensor<string, []>("op_11829_cast")];
            tensor<string, []> var_11831_equation_0 = const()[name = tensor<string, []>("op_11831_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11831_cast = einsum(equation = var_11831_equation_0, values = (var_11689_cast, var_11803_cast))[name = tensor<string, []>("op_11831_cast")];
            tensor<string, []> var_11833_equation_0 = const()[name = tensor<string, []>("op_11833_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11833_cast = einsum(equation = var_11833_equation_0, values = (var_11693_cast, var_11804_cast))[name = tensor<string, []>("op_11833_cast")];
            tensor<string, []> var_11835_equation_0 = const()[name = tensor<string, []>("op_11835_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11835_cast = einsum(equation = var_11835_equation_0, values = (var_11693_cast, var_11805_cast))[name = tensor<string, []>("op_11835_cast")];
            tensor<string, []> var_11837_equation_0 = const()[name = tensor<string, []>("op_11837_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11837_cast = einsum(equation = var_11837_equation_0, values = (var_11697_cast, var_11806_cast))[name = tensor<string, []>("op_11837_cast")];
            tensor<string, []> var_11839_equation_0 = const()[name = tensor<string, []>("op_11839_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11839_cast = einsum(equation = var_11839_equation_0, values = (var_11697_cast, var_11807_cast))[name = tensor<string, []>("op_11839_cast")];
            tensor<string, []> var_11841_equation_0 = const()[name = tensor<string, []>("op_11841_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11841_cast = einsum(equation = var_11841_equation_0, values = (var_11701_cast, var_11808_cast))[name = tensor<string, []>("op_11841_cast")];
            tensor<string, []> var_11843_equation_0 = const()[name = tensor<string, []>("op_11843_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11843_cast = einsum(equation = var_11843_equation_0, values = (var_11701_cast, var_11809_cast))[name = tensor<string, []>("op_11843_cast")];
            tensor<string, []> var_11845_equation_0 = const()[name = tensor<string, []>("op_11845_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11845_cast = einsum(equation = var_11845_equation_0, values = (var_11705_cast, var_11810_cast))[name = tensor<string, []>("op_11845_cast")];
            tensor<string, []> var_11847_equation_0 = const()[name = tensor<string, []>("op_11847_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11847_cast = einsum(equation = var_11847_equation_0, values = (var_11705_cast, var_11811_cast))[name = tensor<string, []>("op_11847_cast")];
            tensor<string, []> var_11849_equation_0 = const()[name = tensor<string, []>("op_11849_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11849_cast = einsum(equation = var_11849_equation_0, values = (var_11709_cast, var_11812_cast))[name = tensor<string, []>("op_11849_cast")];
            tensor<string, []> var_11851_equation_0 = const()[name = tensor<string, []>("op_11851_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11851_cast = einsum(equation = var_11851_equation_0, values = (var_11709_cast, var_11813_cast))[name = tensor<string, []>("op_11851_cast")];
            tensor<string, []> var_11853_equation_0 = const()[name = tensor<string, []>("op_11853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11853_cast = einsum(equation = var_11853_equation_0, values = (var_11713_cast, var_11814_cast))[name = tensor<string, []>("op_11853_cast")];
            tensor<string, []> var_11855_equation_0 = const()[name = tensor<string, []>("op_11855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_11855_cast = einsum(equation = var_11855_equation_0, values = (var_11713_cast, var_11815_cast))[name = tensor<string, []>("op_11855_cast")];
            tensor<bool, []> var_11857_interleave_0 = const()[name = tensor<string, []>("op_11857_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11857_cast = concat(axis = var_10234, interleave = var_11857_interleave_0, values = (var_11817_cast, var_11819_cast))[name = tensor<string, []>("op_11857_cast")];
            tensor<bool, []> var_11859_interleave_0 = const()[name = tensor<string, []>("op_11859_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11859_cast = concat(axis = var_10234, interleave = var_11859_interleave_0, values = (var_11821_cast, var_11823_cast))[name = tensor<string, []>("op_11859_cast")];
            tensor<bool, []> var_11861_interleave_0 = const()[name = tensor<string, []>("op_11861_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11861_cast = concat(axis = var_10234, interleave = var_11861_interleave_0, values = (var_11825_cast, var_11827_cast))[name = tensor<string, []>("op_11861_cast")];
            tensor<bool, []> var_11863_interleave_0 = const()[name = tensor<string, []>("op_11863_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11863_cast = concat(axis = var_10234, interleave = var_11863_interleave_0, values = (var_11829_cast, var_11831_cast))[name = tensor<string, []>("op_11863_cast")];
            tensor<bool, []> var_11865_interleave_0 = const()[name = tensor<string, []>("op_11865_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11865_cast = concat(axis = var_10234, interleave = var_11865_interleave_0, values = (var_11833_cast, var_11835_cast))[name = tensor<string, []>("op_11865_cast")];
            tensor<bool, []> var_11867_interleave_0 = const()[name = tensor<string, []>("op_11867_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11867_cast = concat(axis = var_10234, interleave = var_11867_interleave_0, values = (var_11837_cast, var_11839_cast))[name = tensor<string, []>("op_11867_cast")];
            tensor<bool, []> var_11869_interleave_0 = const()[name = tensor<string, []>("op_11869_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11869_cast = concat(axis = var_10234, interleave = var_11869_interleave_0, values = (var_11841_cast, var_11843_cast))[name = tensor<string, []>("op_11869_cast")];
            tensor<bool, []> var_11871_interleave_0 = const()[name = tensor<string, []>("op_11871_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11871_cast = concat(axis = var_10234, interleave = var_11871_interleave_0, values = (var_11845_cast, var_11847_cast))[name = tensor<string, []>("op_11871_cast")];
            tensor<bool, []> var_11873_interleave_0 = const()[name = tensor<string, []>("op_11873_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11873_cast = concat(axis = var_10234, interleave = var_11873_interleave_0, values = (var_11849_cast, var_11851_cast))[name = tensor<string, []>("op_11873_cast")];
            tensor<bool, []> var_11875_interleave_0 = const()[name = tensor<string, []>("op_11875_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_11875_cast = concat(axis = var_10234, interleave = var_11875_interleave_0, values = (var_11853_cast, var_11855_cast))[name = tensor<string, []>("op_11875_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_10257, interleave = input_411_interleave_0, values = (var_11857_cast, var_11859_cast, var_11861_cast, var_11863_cast, var_11865_cast, var_11867_cast, var_11869_cast, var_11871_cast, var_11873_cast, var_11875_cast))[name = tensor<string, []>("input_411_cast")];
            tensor<int32, [2]> var_11881 = const()[name = tensor<string, []>("op_11881"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11883 = const()[name = tensor<string, []>("op_11883"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11885_pad_type_0 = const()[name = tensor<string, []>("op_11885_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11885_pad_0 = const()[name = tensor<string, []>("op_11885_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204785920))), lut = tensor<fp16, [4]>([-0x1.a18p-6, -0x1.e18p-8, 0x1.ed8p-8, 0x1.a44p-6]), 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, []>(204888384)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11885_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_11883, groups = var_10257, pad = var_11885_pad_0, pad_type = var_11885_pad_type_0, strides = var_11881, 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_11885_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_71_cast = add(x = var_11885_cast, y = inputs_69_cast)[name = tensor<string, []>("inputs_71_cast")];
            tensor<int32, [1]> var_11889 = const()[name = tensor<string, []>("op_11889"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_71_cast = reduce_mean(axes = var_11889, keep_dims = var_10252, 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_11893 = const()[name = tensor<string, []>("op_11893"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_11894_cast = reduce_mean(axes = var_11893, keep_dims = var_10252, x = zero_mean_sq_71_cast)[name = tensor<string, []>("op_11894_cast")];
            tensor<fp16, []> var_11895_to_fp16 = const()[name = tensor<string, []>("op_11895_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_11896_cast = add(x = var_11894_cast, y = var_11895_to_fp16)[name = tensor<string, []>("op_11896_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_11896_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_11900_to_fp16 = const()[name = tensor<string, []>("op_11900_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204889728)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11901_cast = add(x = out_71_cast, y = var_11900_to_fp16)[name = tensor<string, []>("op_11901_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_11903_to_fp16 = const()[name = tensor<string, []>("op_11903_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204891072)))];
            tensor<fp16, [2, 640, 1, 1024]> input_413_cast = mul(x = var_11901_cast, y = var_11903_to_fp16)[name = tensor<string, []>("input_413_cast")];
            tensor<int32, [2]> var_11911 = const()[name = tensor<string, []>("op_11911"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11913 = const()[name = tensor<string, []>("op_11913"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11915_pad_type_0 = const()[name = tensor<string, []>("op_11915_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11915_pad_0 = const()[name = tensor<string, []>("op_11915_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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(204892416))), lut = tensor<fp16, [4]>([-0x1.914p-4, -0x1.d38p-6, 0x1.cbp-6, 0x1.8ep-4]), 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205711680))), lut = tensor<fp16, [4]>([0x1.fep-7, -0x1.4e8p-3, -0x1.3ccp-5, 0x1.3d4p-4]), 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_11915_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_11913, groups = var_10257, pad = var_11915_pad_0, pad_type = var_11915_pad_type_0, strides = var_11911, 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_11915_cast")];
            tensor<int32, [2]> var_11916_split_sizes_0 = const()[name = tensor<string, []>("op_11916_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_11916_axis_0 = const()[name = tensor<string, []>("op_11916_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_11916_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_11916_cast_1 = split(axis = var_11916_axis_0, split_sizes = var_11916_split_sizes_0, x = var_11915_cast)[name = tensor<string, []>("op_11916_cast")];
            tensor<string, []> var_11918_mode_0 = const()[name = tensor<string, []>("op_11918_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_11918_cast = gelu(mode = var_11918_mode_0, x = var_11916_cast_1)[name = tensor<string, []>("op_11918_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_415_cast = mul(x = var_11916_cast_0, y = var_11918_cast)[name = tensor<string, []>("input_415_cast")];
            tensor<int32, [2]> var_11922 = const()[name = tensor<string, []>("op_11922"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11924 = const()[name = tensor<string, []>("op_11924"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_11926_pad_type_0 = const()[name = tensor<string, []>("op_11926_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_11926_pad_0 = const()[name = tensor<string, []>("op_11926_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205713024))), lut = tensor<fp16, [4]>([-0x1.968p-4, -0x1.e1p-6, 0x1.e3cp-6, 0x1.974p-4]), 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, []>(206122688)))];
            tensor<fp16, [2, 640, 1, 1024]> var_11926_cast = conv(bias = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_11924, groups = var_10257, pad = var_11926_pad_0, pad_type = var_11926_pad_type_0, strides = var_11922, weight = up_blocks_2_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_415_cast)[name = tensor<string, []>("op_11926_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_253_cast = add(x = var_11926_cast, y = inputs_71_cast)[name = tensor<string, []>("hidden_states_253_cast")];
            tensor<int32, [4]> var_11928 = const()[name = tensor<string, []>("op_11928"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_417_cast = reshape(shape = var_11928, x = hidden_states_253_cast)[name = tensor<string, []>("input_417_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_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206124032))), lut = tensor<fp16, [4]>([-0x1.ebp-4, -0x1.268p-5, 0x1.1e4p-5, 0x1.e4cp-4]), 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, []>(206226496)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_255_cast = conv(bias = up_blocks_2_attentions_1_proj_out_bias_to_fp16, dilations = var_11934, groups = var_10257, pad = hidden_states_255_pad_0, pad_type = hidden_states_255_pad_type_0, strides = var_11932, 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_10257, 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, []>(206227840)))];
            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, []>(206229824)))];
            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, []>(206231808)))];
            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, []>(206233792)))];
            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_11952 = const()[name = tensor<string, []>("op_11952"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11954 = const()[name = tensor<string, []>("op_11954"), 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, [1382400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206235776))), lut = tensor<fp16, [4]>([-0x1.9ep-4, -0x1.704p-6, 0x1.d3cp-6, 0x1.d44p-4]), 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, []>(207618240)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_259_cast = conv(bias = up_blocks_2_resnets_2_conv1_bias_to_fp16, dilations = var_11954, groups = var_10257, pad = hidden_states_259_pad_0, pad_type = hidden_states_259_pad_type_0, strides = var_11952, 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_11960 = const()[name = tensor<string, []>("op_11960"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11962 = const()[name = tensor<string, []>("op_11962"), 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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207619584))), lut = tensor<fp16, [4]>([-0x1.598p-5, -0x1.69p-8, 0x1.804p-11, 0x1.f24p-8]), 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, []>(207824448)))];
            tensor<fp16, [2, 640, 1, 1]> temb_37_cast = conv(bias = up_blocks_2_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_11962, groups = var_10257, pad = temb_37_pad_0, pad_type = temb_37_pad_type_0, strides = var_11960, 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, []>(207825792)))];
            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, []>(207827136)))];
            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_11972 = const()[name = tensor<string, []>("op_11972"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11974 = const()[name = tensor<string, []>("op_11974"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207828480))), lut = tensor<fp16, [4]>([-0x1.b8cp-4, -0x1.8b8p-6, 0x1.8dp-6, 0x1.bap-4]), 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, []>(208750144)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_261_cast = conv(bias = up_blocks_2_resnets_2_conv2_bias_to_fp16, dilations = var_11974, groups = var_10257, pad = hidden_states_261_pad_0, pad_type = hidden_states_261_pad_type_0, strides = var_11972, 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_11979 = const()[name = tensor<string, []>("op_11979"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_11981 = const()[name = tensor<string, []>("op_11981"), 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, [153600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208751488))), lut = tensor<fp16, [4]>([-0x1.d5p-6, -0x1.0fcp-7, 0x1.128p-7, 0x1.d78p-6]), 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, []>(208905152)))];
            tensor<fp16, [2, 640, 32, 32]> x_21_cast = conv(bias = up_blocks_2_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_11981, groups = var_10257, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = var_11979, 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, []>(208906496)))];
            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, []>(208907840)))];
            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_12001 = const()[name = tensor<string, []>("op_12001"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12003 = const()[name = tensor<string, []>("op_12003"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(208909184))), lut = tensor<fp16, [4]>([-0x1.9bp-4, -0x1.eb4p-6, 0x1.e9p-6, 0x1.998p-4]), 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, []>(209011648)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_265_cast = conv(bias = up_blocks_2_attentions_2_proj_in_bias_to_fp16, dilations = var_12003, groups = var_10257, pad = hidden_states_265_pad_0, pad_type = hidden_states_265_pad_type_0, strides = var_12001, 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_12008 = const()[name = tensor<string, []>("op_12008"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<fp16, [2, 640, 1, 1024]> inputs_73_cast = reshape(shape = var_12008, x = hidden_states_265_cast)[name = tensor<string, []>("inputs_73_cast")];
            tensor<int32, [1]> var_12018 = const()[name = tensor<string, []>("op_12018"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_73_cast = reduce_mean(axes = var_12018, keep_dims = var_10252, 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_12022 = const()[name = tensor<string, []>("op_12022"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_12023_cast = reduce_mean(axes = var_12022, keep_dims = var_10252, x = zero_mean_sq_73_cast)[name = tensor<string, []>("op_12023_cast")];
            tensor<fp16, []> var_12024_to_fp16 = const()[name = tensor<string, []>("op_12024_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_12025_cast = add(x = var_12023_cast, y = var_12024_to_fp16)[name = tensor<string, []>("op_12025_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_12025_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_12029_to_fp16 = const()[name = tensor<string, []>("op_12029_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209012992)))];
            tensor<fp16, [2, 640, 1, 1024]> var_12030_cast = add(x = out_73_cast, y = var_12029_to_fp16)[name = tensor<string, []>("op_12030_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_12032_to_fp16 = const()[name = tensor<string, []>("op_12032_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209014336)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_267_cast = mul(x = var_12030_cast, y = var_12032_to_fp16)[name = tensor<string, []>("hidden_states_267_cast")];
            tensor<int32, [2]> var_12039 = const()[name = tensor<string, []>("op_12039"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12041 = const()[name = tensor<string, []>("op_12041"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209015680))), lut = tensor<fp16, [4]>([-0x1.08p-3, -0x1.3p-5, 0x1.398p-5, 0x1.0b4p-3]), 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_12041, groups = var_10257, pad = q_49_pad_0, pad_type = q_49_pad_type_0, strides = var_12039, 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_12045 = const()[name = tensor<string, []>("op_12045"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12047 = const()[name = tensor<string, []>("op_12047"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209118144))), lut = tensor<fp16, [4]>([-0x1.164p-3, -0x1.33cp-5, 0x1.3c4p-5, 0x1.184p-3]), 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_12047, groups = var_10257, pad = k_97_pad_0, pad_type = k_97_pad_type_0, strides = var_12045, 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_12051 = const()[name = tensor<string, []>("op_12051"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12053 = const()[name = tensor<string, []>("op_12053"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209220608))), lut = tensor<fp16, [4]>([-0x1.6f4p-4, -0x1.a64p-6, 0x1.acp-6, 0x1.724p-4]), 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_12053, groups = var_10257, pad = v_49_pad_0, pad_type = v_49_pad_type_0, strides = var_12051, 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_12057_begin_0 = const()[name = tensor<string, []>("op_12057_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12057_end_0 = const()[name = tensor<string, []>("op_12057_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12057_end_mask_0 = const()[name = tensor<string, []>("op_12057_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12057_cast = slice_by_index(begin = var_12057_begin_0, end = var_12057_end_0, end_mask = var_12057_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12057_cast")];
            tensor<int32, [4]> var_12061_begin_0 = const()[name = tensor<string, []>("op_12061_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12061_end_0 = const()[name = tensor<string, []>("op_12061_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_12061_end_mask_0 = const()[name = tensor<string, []>("op_12061_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12061_cast = slice_by_index(begin = var_12061_begin_0, end = var_12061_end_0, end_mask = var_12061_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12061_cast")];
            tensor<int32, [4]> var_12065_begin_0 = const()[name = tensor<string, []>("op_12065_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12065_end_0 = const()[name = tensor<string, []>("op_12065_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_12065_end_mask_0 = const()[name = tensor<string, []>("op_12065_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12065_cast = slice_by_index(begin = var_12065_begin_0, end = var_12065_end_0, end_mask = var_12065_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12065_cast")];
            tensor<int32, [4]> var_12069_begin_0 = const()[name = tensor<string, []>("op_12069_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12069_end_0 = const()[name = tensor<string, []>("op_12069_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_12069_end_mask_0 = const()[name = tensor<string, []>("op_12069_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12069_cast = slice_by_index(begin = var_12069_begin_0, end = var_12069_end_0, end_mask = var_12069_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12069_cast")];
            tensor<int32, [4]> var_12073_begin_0 = const()[name = tensor<string, []>("op_12073_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12073_end_0 = const()[name = tensor<string, []>("op_12073_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_12073_end_mask_0 = const()[name = tensor<string, []>("op_12073_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12073_cast = slice_by_index(begin = var_12073_begin_0, end = var_12073_end_0, end_mask = var_12073_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12073_cast")];
            tensor<int32, [4]> var_12077_begin_0 = const()[name = tensor<string, []>("op_12077_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12077_end_0 = const()[name = tensor<string, []>("op_12077_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_12077_end_mask_0 = const()[name = tensor<string, []>("op_12077_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12077_cast = slice_by_index(begin = var_12077_begin_0, end = var_12077_end_0, end_mask = var_12077_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12077_cast")];
            tensor<int32, [4]> var_12081_begin_0 = const()[name = tensor<string, []>("op_12081_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12081_end_0 = const()[name = tensor<string, []>("op_12081_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12081_end_mask_0 = const()[name = tensor<string, []>("op_12081_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12081_cast = slice_by_index(begin = var_12081_begin_0, end = var_12081_end_0, end_mask = var_12081_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12081_cast")];
            tensor<int32, [4]> var_12085_begin_0 = const()[name = tensor<string, []>("op_12085_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12085_end_0 = const()[name = tensor<string, []>("op_12085_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12085_end_mask_0 = const()[name = tensor<string, []>("op_12085_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12085_cast = slice_by_index(begin = var_12085_begin_0, end = var_12085_end_0, end_mask = var_12085_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12085_cast")];
            tensor<int32, [4]> var_12089_begin_0 = const()[name = tensor<string, []>("op_12089_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12089_end_0 = const()[name = tensor<string, []>("op_12089_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12089_end_mask_0 = const()[name = tensor<string, []>("op_12089_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12089_cast = slice_by_index(begin = var_12089_begin_0, end = var_12089_end_0, end_mask = var_12089_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12089_cast")];
            tensor<int32, [4]> var_12093_begin_0 = const()[name = tensor<string, []>("op_12093_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12093_end_0 = const()[name = tensor<string, []>("op_12093_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12093_end_mask_0 = const()[name = tensor<string, []>("op_12093_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12093_cast = slice_by_index(begin = var_12093_begin_0, end = var_12093_end_0, end_mask = var_12093_end_mask_0, x = q_49_cast)[name = tensor<string, []>("op_12093_cast")];
            tensor<int32, [4]> var_12096_begin_0 = const()[name = tensor<string, []>("op_12096_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12096_end_0 = const()[name = tensor<string, []>("op_12096_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12096_end_mask_0 = const()[name = tensor<string, []>("op_12096_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12096_cast = slice_by_index(begin = var_12096_begin_0, end = var_12096_end_0, end_mask = var_12096_end_mask_0, x = var_12057_cast)[name = tensor<string, []>("op_12096_cast")];
            tensor<int32, [4]> var_12097_begin_0 = const()[name = tensor<string, []>("op_12097_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12097_end_0 = const()[name = tensor<string, []>("op_12097_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12097_end_mask_0 = const()[name = tensor<string, []>("op_12097_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12097_cast = slice_by_index(begin = var_12097_begin_0, end = var_12097_end_0, end_mask = var_12097_end_mask_0, x = var_12057_cast)[name = tensor<string, []>("op_12097_cast")];
            tensor<int32, [4]> var_12098_begin_0 = const()[name = tensor<string, []>("op_12098_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12098_end_0 = const()[name = tensor<string, []>("op_12098_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12098_end_mask_0 = const()[name = tensor<string, []>("op_12098_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12098_cast = slice_by_index(begin = var_12098_begin_0, end = var_12098_end_0, end_mask = var_12098_end_mask_0, x = var_12061_cast)[name = tensor<string, []>("op_12098_cast")];
            tensor<int32, [4]> var_12099_begin_0 = const()[name = tensor<string, []>("op_12099_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12099_end_0 = const()[name = tensor<string, []>("op_12099_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 1, 512]> var_12099_cast = slice_by_index(begin = var_12099_begin_0, end = var_12099_end_0, end_mask = var_12099_end_mask_0, x = var_12061_cast)[name = tensor<string, []>("op_12099_cast")];
            tensor<int32, [4]> var_12100_begin_0 = const()[name = tensor<string, []>("op_12100_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12100_end_0 = const()[name = tensor<string, []>("op_12100_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12100_end_mask_0 = const()[name = tensor<string, []>("op_12100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12100_cast = slice_by_index(begin = var_12100_begin_0, end = var_12100_end_0, end_mask = var_12100_end_mask_0, x = var_12065_cast)[name = tensor<string, []>("op_12100_cast")];
            tensor<int32, [4]> var_12101_begin_0 = const()[name = tensor<string, []>("op_12101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12101_end_0 = const()[name = tensor<string, []>("op_12101_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12101_end_mask_0 = const()[name = tensor<string, []>("op_12101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12101_cast = slice_by_index(begin = var_12101_begin_0, end = var_12101_end_0, end_mask = var_12101_end_mask_0, x = var_12065_cast)[name = tensor<string, []>("op_12101_cast")];
            tensor<int32, [4]> var_12102_begin_0 = const()[name = tensor<string, []>("op_12102_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12102_end_0 = const()[name = tensor<string, []>("op_12102_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12102_end_mask_0 = const()[name = tensor<string, []>("op_12102_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12102_cast = slice_by_index(begin = var_12102_begin_0, end = var_12102_end_0, end_mask = var_12102_end_mask_0, x = var_12069_cast)[name = tensor<string, []>("op_12102_cast")];
            tensor<int32, [4]> var_12103_begin_0 = const()[name = tensor<string, []>("op_12103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12103_end_0 = const()[name = tensor<string, []>("op_12103_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 1, 512]> var_12103_cast = slice_by_index(begin = var_12103_begin_0, end = var_12103_end_0, end_mask = var_12103_end_mask_0, x = var_12069_cast)[name = tensor<string, []>("op_12103_cast")];
            tensor<int32, [4]> var_12104_begin_0 = const()[name = tensor<string, []>("op_12104_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12104_end_0 = const()[name = tensor<string, []>("op_12104_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12104_end_mask_0 = const()[name = tensor<string, []>("op_12104_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12104_cast = slice_by_index(begin = var_12104_begin_0, end = var_12104_end_0, end_mask = var_12104_end_mask_0, x = var_12073_cast)[name = tensor<string, []>("op_12104_cast")];
            tensor<int32, [4]> var_12105_begin_0 = const()[name = tensor<string, []>("op_12105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12105_end_0 = const()[name = tensor<string, []>("op_12105_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12105_end_mask_0 = const()[name = tensor<string, []>("op_12105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12105_cast = slice_by_index(begin = var_12105_begin_0, end = var_12105_end_0, end_mask = var_12105_end_mask_0, x = var_12073_cast)[name = tensor<string, []>("op_12105_cast")];
            tensor<int32, [4]> var_12106_begin_0 = const()[name = tensor<string, []>("op_12106_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12106_end_0 = const()[name = tensor<string, []>("op_12106_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12106_end_mask_0 = const()[name = tensor<string, []>("op_12106_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12106_cast = slice_by_index(begin = var_12106_begin_0, end = var_12106_end_0, end_mask = var_12106_end_mask_0, x = var_12077_cast)[name = tensor<string, []>("op_12106_cast")];
            tensor<int32, [4]> var_12107_begin_0 = const()[name = tensor<string, []>("op_12107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12107_end_0 = const()[name = tensor<string, []>("op_12107_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 1, 512]> var_12107_cast = slice_by_index(begin = var_12107_begin_0, end = var_12107_end_0, end_mask = var_12107_end_mask_0, x = var_12077_cast)[name = tensor<string, []>("op_12107_cast")];
            tensor<int32, [4]> var_12108_begin_0 = const()[name = tensor<string, []>("op_12108_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12108_end_0 = const()[name = tensor<string, []>("op_12108_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12108_end_mask_0 = const()[name = tensor<string, []>("op_12108_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12108_cast = slice_by_index(begin = var_12108_begin_0, end = var_12108_end_0, end_mask = var_12108_end_mask_0, x = var_12081_cast)[name = tensor<string, []>("op_12108_cast")];
            tensor<int32, [4]> var_12109_begin_0 = const()[name = tensor<string, []>("op_12109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12109_end_0 = const()[name = tensor<string, []>("op_12109_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12109_end_mask_0 = const()[name = tensor<string, []>("op_12109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12109_cast = slice_by_index(begin = var_12109_begin_0, end = var_12109_end_0, end_mask = var_12109_end_mask_0, x = var_12081_cast)[name = tensor<string, []>("op_12109_cast")];
            tensor<int32, [4]> var_12110_begin_0 = const()[name = tensor<string, []>("op_12110_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12110_end_0 = const()[name = tensor<string, []>("op_12110_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12110_end_mask_0 = const()[name = tensor<string, []>("op_12110_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12110_cast = slice_by_index(begin = var_12110_begin_0, end = var_12110_end_0, end_mask = var_12110_end_mask_0, x = var_12085_cast)[name = tensor<string, []>("op_12110_cast")];
            tensor<int32, [4]> var_12111_begin_0 = const()[name = tensor<string, []>("op_12111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12111_end_0 = const()[name = tensor<string, []>("op_12111_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 1, 512]> var_12111_cast = slice_by_index(begin = var_12111_begin_0, end = var_12111_end_0, end_mask = var_12111_end_mask_0, x = var_12085_cast)[name = tensor<string, []>("op_12111_cast")];
            tensor<int32, [4]> var_12112_begin_0 = const()[name = tensor<string, []>("op_12112_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12112_end_0 = const()[name = tensor<string, []>("op_12112_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12112_end_mask_0 = const()[name = tensor<string, []>("op_12112_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12112_cast = slice_by_index(begin = var_12112_begin_0, end = var_12112_end_0, end_mask = var_12112_end_mask_0, x = var_12089_cast)[name = tensor<string, []>("op_12112_cast")];
            tensor<int32, [4]> var_12113_begin_0 = const()[name = tensor<string, []>("op_12113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12113_end_0 = const()[name = tensor<string, []>("op_12113_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12113_end_mask_0 = const()[name = tensor<string, []>("op_12113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12113_cast = slice_by_index(begin = var_12113_begin_0, end = var_12113_end_0, end_mask = var_12113_end_mask_0, x = var_12089_cast)[name = tensor<string, []>("op_12113_cast")];
            tensor<int32, [4]> var_12114_begin_0 = const()[name = tensor<string, []>("op_12114_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12114_end_0 = const()[name = tensor<string, []>("op_12114_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12114_end_mask_0 = const()[name = tensor<string, []>("op_12114_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12114_cast = slice_by_index(begin = var_12114_begin_0, end = var_12114_end_0, end_mask = var_12114_end_mask_0, x = var_12093_cast)[name = tensor<string, []>("op_12114_cast")];
            tensor<int32, [4]> var_12115_begin_0 = const()[name = tensor<string, []>("op_12115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12115_end_0 = const()[name = tensor<string, []>("op_12115_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            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, 64, 1, 512]> var_12115_cast = slice_by_index(begin = var_12115_begin_0, end = var_12115_end_0, end_mask = var_12115_end_mask_0, x = var_12093_cast)[name = tensor<string, []>("op_12115_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_12120_begin_0 = const()[name = tensor<string, []>("op_12120_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12120_end_0 = const()[name = tensor<string, []>("op_12120_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 64])];
            tensor<bool, [4]> var_12120_end_mask_0 = const()[name = tensor<string, []>("op_12120_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, 64]> var_12120_cast = slice_by_index(begin = var_12120_begin_0, end = var_12120_end_0, end_mask = var_12120_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12120_cast")];
            tensor<int32, [4]> var_12124_begin_0 = const()[name = tensor<string, []>("op_12124_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_12124_end_0 = const()[name = tensor<string, []>("op_12124_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 128])];
            tensor<bool, [4]> var_12124_end_mask_0 = const()[name = tensor<string, []>("op_12124_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12124_cast = slice_by_index(begin = var_12124_begin_0, end = var_12124_end_0, end_mask = var_12124_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12124_cast")];
            tensor<int32, [4]> var_12128_begin_0 = const()[name = tensor<string, []>("op_12128_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_12128_end_0 = const()[name = tensor<string, []>("op_12128_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 192])];
            tensor<bool, [4]> var_12128_end_mask_0 = const()[name = tensor<string, []>("op_12128_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12128_cast = slice_by_index(begin = var_12128_begin_0, end = var_12128_end_0, end_mask = var_12128_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12128_cast")];
            tensor<int32, [4]> var_12132_begin_0 = const()[name = tensor<string, []>("op_12132_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_12132_end_0 = const()[name = tensor<string, []>("op_12132_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 256])];
            tensor<bool, [4]> var_12132_end_mask_0 = const()[name = tensor<string, []>("op_12132_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12132_cast = slice_by_index(begin = var_12132_begin_0, end = var_12132_end_0, end_mask = var_12132_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12132_cast")];
            tensor<int32, [4]> var_12136_begin_0 = const()[name = tensor<string, []>("op_12136_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_12136_end_0 = const()[name = tensor<string, []>("op_12136_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 320])];
            tensor<bool, [4]> var_12136_end_mask_0 = const()[name = tensor<string, []>("op_12136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12136_cast = slice_by_index(begin = var_12136_begin_0, end = var_12136_end_0, end_mask = var_12136_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12136_cast")];
            tensor<int32, [4]> var_12140_begin_0 = const()[name = tensor<string, []>("op_12140_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_12140_end_0 = const()[name = tensor<string, []>("op_12140_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 384])];
            tensor<bool, [4]> var_12140_end_mask_0 = const()[name = tensor<string, []>("op_12140_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12140_cast = slice_by_index(begin = var_12140_begin_0, end = var_12140_end_0, end_mask = var_12140_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12140_cast")];
            tensor<int32, [4]> var_12144_begin_0 = const()[name = tensor<string, []>("op_12144_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_12144_end_0 = const()[name = tensor<string, []>("op_12144_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 448])];
            tensor<bool, [4]> var_12144_end_mask_0 = const()[name = tensor<string, []>("op_12144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12144_cast = slice_by_index(begin = var_12144_begin_0, end = var_12144_end_0, end_mask = var_12144_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12144_cast")];
            tensor<int32, [4]> var_12148_begin_0 = const()[name = tensor<string, []>("op_12148_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_12148_end_0 = const()[name = tensor<string, []>("op_12148_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 512])];
            tensor<bool, [4]> var_12148_end_mask_0 = const()[name = tensor<string, []>("op_12148_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12148_cast = slice_by_index(begin = var_12148_begin_0, end = var_12148_end_0, end_mask = var_12148_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12148_cast")];
            tensor<int32, [4]> var_12152_begin_0 = const()[name = tensor<string, []>("op_12152_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12152_end_0 = const()[name = tensor<string, []>("op_12152_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 576])];
            tensor<bool, [4]> var_12152_end_mask_0 = const()[name = tensor<string, []>("op_12152_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12152_cast = slice_by_index(begin = var_12152_begin_0, end = var_12152_end_0, end_mask = var_12152_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12152_cast")];
            tensor<int32, [4]> var_12156_begin_0 = const()[name = tensor<string, []>("op_12156_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_12156_end_0 = const()[name = tensor<string, []>("op_12156_end_0"), val = tensor<int32, [4]>([2, 1024, 1, 640])];
            tensor<bool, [4]> var_12156_end_mask_0 = const()[name = tensor<string, []>("op_12156_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 1024, 1, 64]> var_12156_cast = slice_by_index(begin = var_12156_begin_0, end = var_12156_end_0, end_mask = var_12156_end_mask_0, x = transpose_7)[name = tensor<string, []>("op_12156_cast")];
            tensor<int32, [4]> var_12158_begin_0 = const()[name = tensor<string, []>("op_12158_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12158_end_0 = const()[name = tensor<string, []>("op_12158_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12158_end_mask_0 = const()[name = tensor<string, []>("op_12158_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12158_cast = slice_by_index(begin = var_12158_begin_0, end = var_12158_end_0, end_mask = var_12158_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12158_cast")];
            tensor<int32, [4]> var_12162_begin_0 = const()[name = tensor<string, []>("op_12162_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12162_end_0 = const()[name = tensor<string, []>("op_12162_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_12162_end_mask_0 = const()[name = tensor<string, []>("op_12162_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12162_cast = slice_by_index(begin = var_12162_begin_0, end = var_12162_end_0, end_mask = var_12162_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12162_cast")];
            tensor<int32, [4]> var_12166_begin_0 = const()[name = tensor<string, []>("op_12166_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12166_end_0 = const()[name = tensor<string, []>("op_12166_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_12166_end_mask_0 = const()[name = tensor<string, []>("op_12166_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12166_cast = slice_by_index(begin = var_12166_begin_0, end = var_12166_end_0, end_mask = var_12166_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12166_cast")];
            tensor<int32, [4]> var_12170_begin_0 = const()[name = tensor<string, []>("op_12170_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12170_end_0 = const()[name = tensor<string, []>("op_12170_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_12170_end_mask_0 = const()[name = tensor<string, []>("op_12170_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12170_cast = slice_by_index(begin = var_12170_begin_0, end = var_12170_end_0, end_mask = var_12170_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12170_cast")];
            tensor<int32, [4]> var_12174_begin_0 = const()[name = tensor<string, []>("op_12174_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12174_end_0 = const()[name = tensor<string, []>("op_12174_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_12174_end_mask_0 = const()[name = tensor<string, []>("op_12174_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12174_cast = slice_by_index(begin = var_12174_begin_0, end = var_12174_end_0, end_mask = var_12174_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12174_cast")];
            tensor<int32, [4]> var_12178_begin_0 = const()[name = tensor<string, []>("op_12178_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12178_end_0 = const()[name = tensor<string, []>("op_12178_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_12178_end_mask_0 = const()[name = tensor<string, []>("op_12178_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12178_cast = slice_by_index(begin = var_12178_begin_0, end = var_12178_end_0, end_mask = var_12178_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12178_cast")];
            tensor<int32, [4]> var_12182_begin_0 = const()[name = tensor<string, []>("op_12182_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12182_end_0 = const()[name = tensor<string, []>("op_12182_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12182_end_mask_0 = const()[name = tensor<string, []>("op_12182_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12182_cast = slice_by_index(begin = var_12182_begin_0, end = var_12182_end_0, end_mask = var_12182_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12182_cast")];
            tensor<int32, [4]> var_12186_begin_0 = const()[name = tensor<string, []>("op_12186_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12186_end_0 = const()[name = tensor<string, []>("op_12186_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12186_end_mask_0 = const()[name = tensor<string, []>("op_12186_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12186_cast = slice_by_index(begin = var_12186_begin_0, end = var_12186_end_0, end_mask = var_12186_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12186_cast")];
            tensor<int32, [4]> var_12190_begin_0 = const()[name = tensor<string, []>("op_12190_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12190_end_0 = const()[name = tensor<string, []>("op_12190_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12190_end_mask_0 = const()[name = tensor<string, []>("op_12190_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12190_cast = slice_by_index(begin = var_12190_begin_0, end = var_12190_end_0, end_mask = var_12190_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12190_cast")];
            tensor<int32, [4]> var_12194_begin_0 = const()[name = tensor<string, []>("op_12194_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12194_end_0 = const()[name = tensor<string, []>("op_12194_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12194_end_mask_0 = const()[name = tensor<string, []>("op_12194_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12194_cast = slice_by_index(begin = var_12194_begin_0, end = var_12194_end_0, end_mask = var_12194_end_mask_0, x = v_49_cast)[name = tensor<string, []>("op_12194_cast")];
            tensor<string, []> var_12198_equation_0 = const()[name = tensor<string, []>("op_12198_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12198_cast = einsum(equation = var_12198_equation_0, values = (var_12120_cast, var_12096_cast))[name = tensor<string, []>("op_12198_cast")];
            tensor<fp16, []> var_12199_to_fp16 = const()[name = tensor<string, []>("op_12199_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_641_cast = mul(x = var_12198_cast, y = var_12199_to_fp16)[name = tensor<string, []>("aw_chunk_641_cast")];
            tensor<string, []> var_12202_equation_0 = const()[name = tensor<string, []>("op_12202_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12202_cast = einsum(equation = var_12202_equation_0, values = (var_12120_cast, var_12097_cast))[name = tensor<string, []>("op_12202_cast")];
            tensor<fp16, []> var_12203_to_fp16 = const()[name = tensor<string, []>("op_12203_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_643_cast = mul(x = var_12202_cast, y = var_12203_to_fp16)[name = tensor<string, []>("aw_chunk_643_cast")];
            tensor<string, []> var_12206_equation_0 = const()[name = tensor<string, []>("op_12206_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12206_cast = einsum(equation = var_12206_equation_0, values = (var_12124_cast, var_12098_cast))[name = tensor<string, []>("op_12206_cast")];
            tensor<fp16, []> var_12207_to_fp16 = const()[name = tensor<string, []>("op_12207_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_645_cast = mul(x = var_12206_cast, y = var_12207_to_fp16)[name = tensor<string, []>("aw_chunk_645_cast")];
            tensor<string, []> var_12210_equation_0 = const()[name = tensor<string, []>("op_12210_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12210_cast = einsum(equation = var_12210_equation_0, values = (var_12124_cast, var_12099_cast))[name = tensor<string, []>("op_12210_cast")];
            tensor<fp16, []> var_12211_to_fp16 = const()[name = tensor<string, []>("op_12211_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_647_cast = mul(x = var_12210_cast, y = var_12211_to_fp16)[name = tensor<string, []>("aw_chunk_647_cast")];
            tensor<string, []> var_12214_equation_0 = const()[name = tensor<string, []>("op_12214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12214_cast = einsum(equation = var_12214_equation_0, values = (var_12128_cast, var_12100_cast))[name = tensor<string, []>("op_12214_cast")];
            tensor<fp16, []> var_12215_to_fp16 = const()[name = tensor<string, []>("op_12215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_649_cast = mul(x = var_12214_cast, y = var_12215_to_fp16)[name = tensor<string, []>("aw_chunk_649_cast")];
            tensor<string, []> var_12218_equation_0 = const()[name = tensor<string, []>("op_12218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12218_cast = einsum(equation = var_12218_equation_0, values = (var_12128_cast, var_12101_cast))[name = tensor<string, []>("op_12218_cast")];
            tensor<fp16, []> var_12219_to_fp16 = const()[name = tensor<string, []>("op_12219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_651_cast = mul(x = var_12218_cast, y = var_12219_to_fp16)[name = tensor<string, []>("aw_chunk_651_cast")];
            tensor<string, []> var_12222_equation_0 = const()[name = tensor<string, []>("op_12222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12222_cast = einsum(equation = var_12222_equation_0, values = (var_12132_cast, var_12102_cast))[name = tensor<string, []>("op_12222_cast")];
            tensor<fp16, []> var_12223_to_fp16 = const()[name = tensor<string, []>("op_12223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_653_cast = mul(x = var_12222_cast, y = var_12223_to_fp16)[name = tensor<string, []>("aw_chunk_653_cast")];
            tensor<string, []> var_12226_equation_0 = const()[name = tensor<string, []>("op_12226_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12226_cast = einsum(equation = var_12226_equation_0, values = (var_12132_cast, var_12103_cast))[name = tensor<string, []>("op_12226_cast")];
            tensor<fp16, []> var_12227_to_fp16 = const()[name = tensor<string, []>("op_12227_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_655_cast = mul(x = var_12226_cast, y = var_12227_to_fp16)[name = tensor<string, []>("aw_chunk_655_cast")];
            tensor<string, []> var_12230_equation_0 = const()[name = tensor<string, []>("op_12230_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12230_cast = einsum(equation = var_12230_equation_0, values = (var_12136_cast, var_12104_cast))[name = tensor<string, []>("op_12230_cast")];
            tensor<fp16, []> var_12231_to_fp16 = const()[name = tensor<string, []>("op_12231_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_657_cast = mul(x = var_12230_cast, y = var_12231_to_fp16)[name = tensor<string, []>("aw_chunk_657_cast")];
            tensor<string, []> var_12234_equation_0 = const()[name = tensor<string, []>("op_12234_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12234_cast = einsum(equation = var_12234_equation_0, values = (var_12136_cast, var_12105_cast))[name = tensor<string, []>("op_12234_cast")];
            tensor<fp16, []> var_12235_to_fp16 = const()[name = tensor<string, []>("op_12235_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_659_cast = mul(x = var_12234_cast, y = var_12235_to_fp16)[name = tensor<string, []>("aw_chunk_659_cast")];
            tensor<string, []> var_12238_equation_0 = const()[name = tensor<string, []>("op_12238_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12238_cast = einsum(equation = var_12238_equation_0, values = (var_12140_cast, var_12106_cast))[name = tensor<string, []>("op_12238_cast")];
            tensor<fp16, []> var_12239_to_fp16 = const()[name = tensor<string, []>("op_12239_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_661_cast = mul(x = var_12238_cast, y = var_12239_to_fp16)[name = tensor<string, []>("aw_chunk_661_cast")];
            tensor<string, []> var_12242_equation_0 = const()[name = tensor<string, []>("op_12242_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12242_cast = einsum(equation = var_12242_equation_0, values = (var_12140_cast, var_12107_cast))[name = tensor<string, []>("op_12242_cast")];
            tensor<fp16, []> var_12243_to_fp16 = const()[name = tensor<string, []>("op_12243_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_663_cast = mul(x = var_12242_cast, y = var_12243_to_fp16)[name = tensor<string, []>("aw_chunk_663_cast")];
            tensor<string, []> var_12246_equation_0 = const()[name = tensor<string, []>("op_12246_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12246_cast = einsum(equation = var_12246_equation_0, values = (var_12144_cast, var_12108_cast))[name = tensor<string, []>("op_12246_cast")];
            tensor<fp16, []> var_12247_to_fp16 = const()[name = tensor<string, []>("op_12247_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_665_cast = mul(x = var_12246_cast, y = var_12247_to_fp16)[name = tensor<string, []>("aw_chunk_665_cast")];
            tensor<string, []> var_12250_equation_0 = const()[name = tensor<string, []>("op_12250_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12250_cast = einsum(equation = var_12250_equation_0, values = (var_12144_cast, var_12109_cast))[name = tensor<string, []>("op_12250_cast")];
            tensor<fp16, []> var_12251_to_fp16 = const()[name = tensor<string, []>("op_12251_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_667_cast = mul(x = var_12250_cast, y = var_12251_to_fp16)[name = tensor<string, []>("aw_chunk_667_cast")];
            tensor<string, []> var_12254_equation_0 = const()[name = tensor<string, []>("op_12254_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12254_cast = einsum(equation = var_12254_equation_0, values = (var_12148_cast, var_12110_cast))[name = tensor<string, []>("op_12254_cast")];
            tensor<fp16, []> var_12255_to_fp16 = const()[name = tensor<string, []>("op_12255_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_669_cast = mul(x = var_12254_cast, y = var_12255_to_fp16)[name = tensor<string, []>("aw_chunk_669_cast")];
            tensor<string, []> var_12258_equation_0 = const()[name = tensor<string, []>("op_12258_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12258_cast = einsum(equation = var_12258_equation_0, values = (var_12148_cast, var_12111_cast))[name = tensor<string, []>("op_12258_cast")];
            tensor<fp16, []> var_12259_to_fp16 = const()[name = tensor<string, []>("op_12259_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_671_cast = mul(x = var_12258_cast, y = var_12259_to_fp16)[name = tensor<string, []>("aw_chunk_671_cast")];
            tensor<string, []> var_12262_equation_0 = const()[name = tensor<string, []>("op_12262_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12262_cast = einsum(equation = var_12262_equation_0, values = (var_12152_cast, var_12112_cast))[name = tensor<string, []>("op_12262_cast")];
            tensor<fp16, []> var_12263_to_fp16 = const()[name = tensor<string, []>("op_12263_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_673_cast = mul(x = var_12262_cast, y = var_12263_to_fp16)[name = tensor<string, []>("aw_chunk_673_cast")];
            tensor<string, []> var_12266_equation_0 = const()[name = tensor<string, []>("op_12266_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12266_cast = einsum(equation = var_12266_equation_0, values = (var_12152_cast, var_12113_cast))[name = tensor<string, []>("op_12266_cast")];
            tensor<fp16, []> var_12267_to_fp16 = const()[name = tensor<string, []>("op_12267_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_675_cast = mul(x = var_12266_cast, y = var_12267_to_fp16)[name = tensor<string, []>("aw_chunk_675_cast")];
            tensor<string, []> var_12270_equation_0 = const()[name = tensor<string, []>("op_12270_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12270_cast = einsum(equation = var_12270_equation_0, values = (var_12156_cast, var_12114_cast))[name = tensor<string, []>("op_12270_cast")];
            tensor<fp16, []> var_12271_to_fp16 = const()[name = tensor<string, []>("op_12271_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_677_cast = mul(x = var_12270_cast, y = var_12271_to_fp16)[name = tensor<string, []>("aw_chunk_677_cast")];
            tensor<string, []> var_12274_equation_0 = const()[name = tensor<string, []>("op_12274_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 1024, 1, 512]> var_12274_cast = einsum(equation = var_12274_equation_0, values = (var_12156_cast, var_12115_cast))[name = tensor<string, []>("op_12274_cast")];
            tensor<fp16, []> var_12275_to_fp16 = const()[name = tensor<string, []>("op_12275_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 1024, 1, 512]> aw_chunk_679_cast = mul(x = var_12274_cast, y = var_12275_to_fp16)[name = tensor<string, []>("aw_chunk_679_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12277_cast = softmax(axis = var_10257, x = aw_chunk_641_cast)[name = tensor<string, []>("op_12277_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12278_cast = softmax(axis = var_10257, x = aw_chunk_643_cast)[name = tensor<string, []>("op_12278_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12279_cast = softmax(axis = var_10257, x = aw_chunk_645_cast)[name = tensor<string, []>("op_12279_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12280_cast = softmax(axis = var_10257, x = aw_chunk_647_cast)[name = tensor<string, []>("op_12280_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12281_cast = softmax(axis = var_10257, x = aw_chunk_649_cast)[name = tensor<string, []>("op_12281_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12282_cast = softmax(axis = var_10257, x = aw_chunk_651_cast)[name = tensor<string, []>("op_12282_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12283_cast = softmax(axis = var_10257, x = aw_chunk_653_cast)[name = tensor<string, []>("op_12283_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12284_cast = softmax(axis = var_10257, x = aw_chunk_655_cast)[name = tensor<string, []>("op_12284_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12285_cast = softmax(axis = var_10257, x = aw_chunk_657_cast)[name = tensor<string, []>("op_12285_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12286_cast = softmax(axis = var_10257, x = aw_chunk_659_cast)[name = tensor<string, []>("op_12286_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12287_cast = softmax(axis = var_10257, x = aw_chunk_661_cast)[name = tensor<string, []>("op_12287_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12288_cast = softmax(axis = var_10257, x = aw_chunk_663_cast)[name = tensor<string, []>("op_12288_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12289_cast = softmax(axis = var_10257, x = aw_chunk_665_cast)[name = tensor<string, []>("op_12289_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12290_cast = softmax(axis = var_10257, x = aw_chunk_667_cast)[name = tensor<string, []>("op_12290_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12291_cast = softmax(axis = var_10257, x = aw_chunk_669_cast)[name = tensor<string, []>("op_12291_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12292_cast = softmax(axis = var_10257, x = aw_chunk_671_cast)[name = tensor<string, []>("op_12292_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12293_cast = softmax(axis = var_10257, x = aw_chunk_673_cast)[name = tensor<string, []>("op_12293_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12294_cast = softmax(axis = var_10257, x = aw_chunk_675_cast)[name = tensor<string, []>("op_12294_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12295_cast = softmax(axis = var_10257, x = aw_chunk_677_cast)[name = tensor<string, []>("op_12295_cast")];
            tensor<fp16, [2, 1024, 1, 512]> var_12296_cast = softmax(axis = var_10257, x = aw_chunk_679_cast)[name = tensor<string, []>("op_12296_cast")];
            tensor<string, []> var_12298_equation_0 = const()[name = tensor<string, []>("op_12298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12298_cast = einsum(equation = var_12298_equation_0, values = (var_12158_cast, var_12277_cast))[name = tensor<string, []>("op_12298_cast")];
            tensor<string, []> var_12300_equation_0 = const()[name = tensor<string, []>("op_12300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12300_cast = einsum(equation = var_12300_equation_0, values = (var_12158_cast, var_12278_cast))[name = tensor<string, []>("op_12300_cast")];
            tensor<string, []> var_12302_equation_0 = const()[name = tensor<string, []>("op_12302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12302_cast = einsum(equation = var_12302_equation_0, values = (var_12162_cast, var_12279_cast))[name = tensor<string, []>("op_12302_cast")];
            tensor<string, []> var_12304_equation_0 = const()[name = tensor<string, []>("op_12304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12304_cast = einsum(equation = var_12304_equation_0, values = (var_12162_cast, var_12280_cast))[name = tensor<string, []>("op_12304_cast")];
            tensor<string, []> var_12306_equation_0 = const()[name = tensor<string, []>("op_12306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12306_cast = einsum(equation = var_12306_equation_0, values = (var_12166_cast, var_12281_cast))[name = tensor<string, []>("op_12306_cast")];
            tensor<string, []> var_12308_equation_0 = const()[name = tensor<string, []>("op_12308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12308_cast = einsum(equation = var_12308_equation_0, values = (var_12166_cast, var_12282_cast))[name = tensor<string, []>("op_12308_cast")];
            tensor<string, []> var_12310_equation_0 = const()[name = tensor<string, []>("op_12310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12310_cast = einsum(equation = var_12310_equation_0, values = (var_12170_cast, var_12283_cast))[name = tensor<string, []>("op_12310_cast")];
            tensor<string, []> var_12312_equation_0 = const()[name = tensor<string, []>("op_12312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12312_cast = einsum(equation = var_12312_equation_0, values = (var_12170_cast, var_12284_cast))[name = tensor<string, []>("op_12312_cast")];
            tensor<string, []> var_12314_equation_0 = const()[name = tensor<string, []>("op_12314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12314_cast = einsum(equation = var_12314_equation_0, values = (var_12174_cast, var_12285_cast))[name = tensor<string, []>("op_12314_cast")];
            tensor<string, []> var_12316_equation_0 = const()[name = tensor<string, []>("op_12316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12316_cast = einsum(equation = var_12316_equation_0, values = (var_12174_cast, var_12286_cast))[name = tensor<string, []>("op_12316_cast")];
            tensor<string, []> var_12318_equation_0 = const()[name = tensor<string, []>("op_12318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12318_cast = einsum(equation = var_12318_equation_0, values = (var_12178_cast, var_12287_cast))[name = tensor<string, []>("op_12318_cast")];
            tensor<string, []> var_12320_equation_0 = const()[name = tensor<string, []>("op_12320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12320_cast = einsum(equation = var_12320_equation_0, values = (var_12178_cast, var_12288_cast))[name = tensor<string, []>("op_12320_cast")];
            tensor<string, []> var_12322_equation_0 = const()[name = tensor<string, []>("op_12322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12322_cast = einsum(equation = var_12322_equation_0, values = (var_12182_cast, var_12289_cast))[name = tensor<string, []>("op_12322_cast")];
            tensor<string, []> var_12324_equation_0 = const()[name = tensor<string, []>("op_12324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12324_cast = einsum(equation = var_12324_equation_0, values = (var_12182_cast, var_12290_cast))[name = tensor<string, []>("op_12324_cast")];
            tensor<string, []> var_12326_equation_0 = const()[name = tensor<string, []>("op_12326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12326_cast = einsum(equation = var_12326_equation_0, values = (var_12186_cast, var_12291_cast))[name = tensor<string, []>("op_12326_cast")];
            tensor<string, []> var_12328_equation_0 = const()[name = tensor<string, []>("op_12328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12328_cast = einsum(equation = var_12328_equation_0, values = (var_12186_cast, var_12292_cast))[name = tensor<string, []>("op_12328_cast")];
            tensor<string, []> var_12330_equation_0 = const()[name = tensor<string, []>("op_12330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12330_cast = einsum(equation = var_12330_equation_0, values = (var_12190_cast, var_12293_cast))[name = tensor<string, []>("op_12330_cast")];
            tensor<string, []> var_12332_equation_0 = const()[name = tensor<string, []>("op_12332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12332_cast = einsum(equation = var_12332_equation_0, values = (var_12190_cast, var_12294_cast))[name = tensor<string, []>("op_12332_cast")];
            tensor<string, []> var_12334_equation_0 = const()[name = tensor<string, []>("op_12334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12334_cast = einsum(equation = var_12334_equation_0, values = (var_12194_cast, var_12295_cast))[name = tensor<string, []>("op_12334_cast")];
            tensor<string, []> var_12336_equation_0 = const()[name = tensor<string, []>("op_12336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12336_cast = einsum(equation = var_12336_equation_0, values = (var_12194_cast, var_12296_cast))[name = tensor<string, []>("op_12336_cast")];
            tensor<bool, []> var_12338_interleave_0 = const()[name = tensor<string, []>("op_12338_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12338_cast = concat(axis = var_10234, interleave = var_12338_interleave_0, values = (var_12298_cast, var_12300_cast))[name = tensor<string, []>("op_12338_cast")];
            tensor<bool, []> var_12340_interleave_0 = const()[name = tensor<string, []>("op_12340_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12340_cast = concat(axis = var_10234, interleave = var_12340_interleave_0, values = (var_12302_cast, var_12304_cast))[name = tensor<string, []>("op_12340_cast")];
            tensor<bool, []> var_12342_interleave_0 = const()[name = tensor<string, []>("op_12342_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12342_cast = concat(axis = var_10234, interleave = var_12342_interleave_0, values = (var_12306_cast, var_12308_cast))[name = tensor<string, []>("op_12342_cast")];
            tensor<bool, []> var_12344_interleave_0 = const()[name = tensor<string, []>("op_12344_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12344_cast = concat(axis = var_10234, interleave = var_12344_interleave_0, values = (var_12310_cast, var_12312_cast))[name = tensor<string, []>("op_12344_cast")];
            tensor<bool, []> var_12346_interleave_0 = const()[name = tensor<string, []>("op_12346_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12346_cast = concat(axis = var_10234, interleave = var_12346_interleave_0, values = (var_12314_cast, var_12316_cast))[name = tensor<string, []>("op_12346_cast")];
            tensor<bool, []> var_12348_interleave_0 = const()[name = tensor<string, []>("op_12348_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12348_cast = concat(axis = var_10234, interleave = var_12348_interleave_0, values = (var_12318_cast, var_12320_cast))[name = tensor<string, []>("op_12348_cast")];
            tensor<bool, []> var_12350_interleave_0 = const()[name = tensor<string, []>("op_12350_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12350_cast = concat(axis = var_10234, interleave = var_12350_interleave_0, values = (var_12322_cast, var_12324_cast))[name = tensor<string, []>("op_12350_cast")];
            tensor<bool, []> var_12352_interleave_0 = const()[name = tensor<string, []>("op_12352_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12352_cast = concat(axis = var_10234, interleave = var_12352_interleave_0, values = (var_12326_cast, var_12328_cast))[name = tensor<string, []>("op_12352_cast")];
            tensor<bool, []> var_12354_interleave_0 = const()[name = tensor<string, []>("op_12354_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12354_cast = concat(axis = var_10234, interleave = var_12354_interleave_0, values = (var_12330_cast, var_12332_cast))[name = tensor<string, []>("op_12354_cast")];
            tensor<bool, []> var_12356_interleave_0 = const()[name = tensor<string, []>("op_12356_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12356_cast = concat(axis = var_10234, interleave = var_12356_interleave_0, values = (var_12334_cast, var_12336_cast))[name = tensor<string, []>("op_12356_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_10257, interleave = input_435_interleave_0, values = (var_12338_cast, var_12340_cast, var_12342_cast, var_12344_cast, var_12346_cast, var_12348_cast, var_12350_cast, var_12352_cast, var_12354_cast, var_12356_cast))[name = tensor<string, []>("input_435_cast")];
            tensor<int32, [2]> var_12362 = const()[name = tensor<string, []>("op_12362"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12364 = const()[name = tensor<string, []>("op_12364"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12366_pad_type_0 = const()[name = tensor<string, []>("op_12366_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12366_pad_0 = const()[name = tensor<string, []>("op_12366_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209323072))), lut = tensor<fp16, [4]>([-0x1.6cp-4, -0x1.afp-6, 0x1.a44p-6, 0x1.69p-4]), 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, []>(209425536)))];
            tensor<fp16, [2, 640, 1, 1024]> var_12366_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_12364, groups = var_10257, pad = var_12366_pad_0, pad_type = var_12366_pad_type_0, strides = var_12362, 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_12366_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_75_cast = add(x = var_12366_cast, y = inputs_73_cast)[name = tensor<string, []>("inputs_75_cast")];
            tensor<int32, [1]> var_12370 = const()[name = tensor<string, []>("op_12370"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_75_cast = reduce_mean(axes = var_12370, keep_dims = var_10252, 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_12374 = const()[name = tensor<string, []>("op_12374"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_12375_cast = reduce_mean(axes = var_12374, keep_dims = var_10252, x = zero_mean_sq_75_cast)[name = tensor<string, []>("op_12375_cast")];
            tensor<fp16, []> var_12376_to_fp16 = const()[name = tensor<string, []>("op_12376_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_12377_cast = add(x = var_12375_cast, y = var_12376_to_fp16)[name = tensor<string, []>("op_12377_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_12377_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_12381_to_fp16 = const()[name = tensor<string, []>("op_12381_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209426880)))];
            tensor<fp16, [2, 640, 1, 1024]> var_12382_cast = add(x = out_75_cast, y = var_12381_to_fp16)[name = tensor<string, []>("op_12382_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_12384_to_fp16 = const()[name = tensor<string, []>("op_12384_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209428224)))];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_269_cast = mul(x = var_12382_cast, y = var_12384_to_fp16)[name = tensor<string, []>("hidden_states_269_cast")];
            tensor<int32, [2]> var_12391 = const()[name = tensor<string, []>("op_12391"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12393 = const()[name = tensor<string, []>("op_12393"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209429568))), lut = tensor<fp16, [4]>([-0x1.6ap-4, -0x1.a74p-6, 0x1.b84p-6, 0x1.6dcp-4]), 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_12393, groups = var_10257, pad = q_51_pad_0, pad_type = q_51_pad_type_0, strides = var_12391, 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_12397 = const()[name = tensor<string, []>("op_12397"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12399 = const()[name = tensor<string, []>("op_12399"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_101_pad_type_0 = const()[name = tensor<string, []>("k_101_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_101_pad_0 = const()[name = tensor<string, []>("k_101_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209532032))), lut = tensor<fp16, [4]>([-0x1.468p-4, -0x1.808p-6, 0x1.818p-6, 0x1.458p-4]), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> k_101_cast = conv(dilations = var_12399, groups = var_10257, pad = k_101_pad_0, pad_type = k_101_pad_type_0, strides = var_12397, 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_12403 = const()[name = tensor<string, []>("op_12403"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12405 = const()[name = tensor<string, []>("op_12405"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_51_pad_type_0 = const()[name = tensor<string, []>("v_51_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_51_pad_0 = const()[name = tensor<string, []>("v_51_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [640, 1024, 1, 1]> up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [163840]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209695936))), lut = tensor<fp16, [4]>([-0x1.7a8p-5, -0x1.b6p-7, 0x1.ad4p-7, 0x1.774p-5]), name = tensor<string, []>("up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([640, 1024, 1, 1])];
            tensor<fp16, [2, 640, 1, 77]> v_51_cast = conv(dilations = var_12405, groups = var_10257, pad = v_51_pad_0, pad_type = v_51_pad_type_0, strides = var_12403, 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_12409_begin_0 = const()[name = tensor<string, []>("op_12409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12409_end_0 = const()[name = tensor<string, []>("op_12409_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12409_end_mask_0 = const()[name = tensor<string, []>("op_12409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12409_cast = slice_by_index(begin = var_12409_begin_0, end = var_12409_end_0, end_mask = var_12409_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12409_cast")];
            tensor<int32, [4]> var_12413_begin_0 = const()[name = tensor<string, []>("op_12413_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12413_end_0 = const()[name = tensor<string, []>("op_12413_end_0"), val = tensor<int32, [4]>([2, 128, 1, 1024])];
            tensor<bool, [4]> var_12413_end_mask_0 = const()[name = tensor<string, []>("op_12413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12413_cast = slice_by_index(begin = var_12413_begin_0, end = var_12413_end_0, end_mask = var_12413_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12413_cast")];
            tensor<int32, [4]> var_12417_begin_0 = const()[name = tensor<string, []>("op_12417_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12417_end_0 = const()[name = tensor<string, []>("op_12417_end_0"), val = tensor<int32, [4]>([2, 192, 1, 1024])];
            tensor<bool, [4]> var_12417_end_mask_0 = const()[name = tensor<string, []>("op_12417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12417_cast = slice_by_index(begin = var_12417_begin_0, end = var_12417_end_0, end_mask = var_12417_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12417_cast")];
            tensor<int32, [4]> var_12421_begin_0 = const()[name = tensor<string, []>("op_12421_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12421_end_0 = const()[name = tensor<string, []>("op_12421_end_0"), val = tensor<int32, [4]>([2, 256, 1, 1024])];
            tensor<bool, [4]> var_12421_end_mask_0 = const()[name = tensor<string, []>("op_12421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12421_cast = slice_by_index(begin = var_12421_begin_0, end = var_12421_end_0, end_mask = var_12421_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12421_cast")];
            tensor<int32, [4]> var_12425_begin_0 = const()[name = tensor<string, []>("op_12425_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12425_end_0 = const()[name = tensor<string, []>("op_12425_end_0"), val = tensor<int32, [4]>([2, 320, 1, 1024])];
            tensor<bool, [4]> var_12425_end_mask_0 = const()[name = tensor<string, []>("op_12425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12425_cast = slice_by_index(begin = var_12425_begin_0, end = var_12425_end_0, end_mask = var_12425_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12425_cast")];
            tensor<int32, [4]> var_12429_begin_0 = const()[name = tensor<string, []>("op_12429_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12429_end_0 = const()[name = tensor<string, []>("op_12429_end_0"), val = tensor<int32, [4]>([2, 384, 1, 1024])];
            tensor<bool, [4]> var_12429_end_mask_0 = const()[name = tensor<string, []>("op_12429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12429_cast = slice_by_index(begin = var_12429_begin_0, end = var_12429_end_0, end_mask = var_12429_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12429_cast")];
            tensor<int32, [4]> var_12433_begin_0 = const()[name = tensor<string, []>("op_12433_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12433_end_0 = const()[name = tensor<string, []>("op_12433_end_0"), val = tensor<int32, [4]>([2, 448, 1, 1024])];
            tensor<bool, [4]> var_12433_end_mask_0 = const()[name = tensor<string, []>("op_12433_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12433_cast = slice_by_index(begin = var_12433_begin_0, end = var_12433_end_0, end_mask = var_12433_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12433_cast")];
            tensor<int32, [4]> var_12437_begin_0 = const()[name = tensor<string, []>("op_12437_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12437_end_0 = const()[name = tensor<string, []>("op_12437_end_0"), val = tensor<int32, [4]>([2, 512, 1, 1024])];
            tensor<bool, [4]> var_12437_end_mask_0 = const()[name = tensor<string, []>("op_12437_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12437_cast = slice_by_index(begin = var_12437_begin_0, end = var_12437_end_0, end_mask = var_12437_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12437_cast")];
            tensor<int32, [4]> var_12441_begin_0 = const()[name = tensor<string, []>("op_12441_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12441_end_0 = const()[name = tensor<string, []>("op_12441_end_0"), val = tensor<int32, [4]>([2, 576, 1, 1024])];
            tensor<bool, [4]> var_12441_end_mask_0 = const()[name = tensor<string, []>("op_12441_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12441_cast = slice_by_index(begin = var_12441_begin_0, end = var_12441_end_0, end_mask = var_12441_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12441_cast")];
            tensor<int32, [4]> var_12445_begin_0 = const()[name = tensor<string, []>("op_12445_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12445_end_0 = const()[name = tensor<string, []>("op_12445_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1024])];
            tensor<bool, [4]> var_12445_end_mask_0 = const()[name = tensor<string, []>("op_12445_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 1024]> var_12445_cast = slice_by_index(begin = var_12445_begin_0, end = var_12445_end_0, end_mask = var_12445_end_mask_0, x = q_51_cast)[name = tensor<string, []>("op_12445_cast")];
            tensor<int32, [4]> var_12448_begin_0 = const()[name = tensor<string, []>("op_12448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12448_end_0 = const()[name = tensor<string, []>("op_12448_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12448_end_mask_0 = const()[name = tensor<string, []>("op_12448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12448_cast = slice_by_index(begin = var_12448_begin_0, end = var_12448_end_0, end_mask = var_12448_end_mask_0, x = var_12409_cast)[name = tensor<string, []>("op_12448_cast")];
            tensor<int32, [4]> var_12449_begin_0 = const()[name = tensor<string, []>("op_12449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12449_end_0 = const()[name = tensor<string, []>("op_12449_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12449_end_mask_0 = const()[name = tensor<string, []>("op_12449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12449_cast = slice_by_index(begin = var_12449_begin_0, end = var_12449_end_0, end_mask = var_12449_end_mask_0, x = var_12409_cast)[name = tensor<string, []>("op_12449_cast")];
            tensor<int32, [4]> var_12450_begin_0 = const()[name = tensor<string, []>("op_12450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12450_end_0 = const()[name = tensor<string, []>("op_12450_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12450_end_mask_0 = const()[name = tensor<string, []>("op_12450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12450_cast = slice_by_index(begin = var_12450_begin_0, end = var_12450_end_0, end_mask = var_12450_end_mask_0, x = var_12413_cast)[name = tensor<string, []>("op_12450_cast")];
            tensor<int32, [4]> var_12451_begin_0 = const()[name = tensor<string, []>("op_12451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12451_end_0 = const()[name = tensor<string, []>("op_12451_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12451_end_mask_0 = const()[name = tensor<string, []>("op_12451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12451_cast = slice_by_index(begin = var_12451_begin_0, end = var_12451_end_0, end_mask = var_12451_end_mask_0, x = var_12413_cast)[name = tensor<string, []>("op_12451_cast")];
            tensor<int32, [4]> var_12452_begin_0 = const()[name = tensor<string, []>("op_12452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12452_end_0 = const()[name = tensor<string, []>("op_12452_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12452_end_mask_0 = const()[name = tensor<string, []>("op_12452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12452_cast = slice_by_index(begin = var_12452_begin_0, end = var_12452_end_0, end_mask = var_12452_end_mask_0, x = var_12417_cast)[name = tensor<string, []>("op_12452_cast")];
            tensor<int32, [4]> var_12453_begin_0 = const()[name = tensor<string, []>("op_12453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12453_end_0 = const()[name = tensor<string, []>("op_12453_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12453_end_mask_0 = const()[name = tensor<string, []>("op_12453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12453_cast = slice_by_index(begin = var_12453_begin_0, end = var_12453_end_0, end_mask = var_12453_end_mask_0, x = var_12417_cast)[name = tensor<string, []>("op_12453_cast")];
            tensor<int32, [4]> var_12454_begin_0 = const()[name = tensor<string, []>("op_12454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12454_end_0 = const()[name = tensor<string, []>("op_12454_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12454_end_mask_0 = const()[name = tensor<string, []>("op_12454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12454_cast = slice_by_index(begin = var_12454_begin_0, end = var_12454_end_0, end_mask = var_12454_end_mask_0, x = var_12421_cast)[name = tensor<string, []>("op_12454_cast")];
            tensor<int32, [4]> var_12455_begin_0 = const()[name = tensor<string, []>("op_12455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12455_end_0 = const()[name = tensor<string, []>("op_12455_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12455_end_mask_0 = const()[name = tensor<string, []>("op_12455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12455_cast = slice_by_index(begin = var_12455_begin_0, end = var_12455_end_0, end_mask = var_12455_end_mask_0, x = var_12421_cast)[name = tensor<string, []>("op_12455_cast")];
            tensor<int32, [4]> var_12456_begin_0 = const()[name = tensor<string, []>("op_12456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12456_end_0 = const()[name = tensor<string, []>("op_12456_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12456_end_mask_0 = const()[name = tensor<string, []>("op_12456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12456_cast = slice_by_index(begin = var_12456_begin_0, end = var_12456_end_0, end_mask = var_12456_end_mask_0, x = var_12425_cast)[name = tensor<string, []>("op_12456_cast")];
            tensor<int32, [4]> var_12457_begin_0 = const()[name = tensor<string, []>("op_12457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12457_end_0 = const()[name = tensor<string, []>("op_12457_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12457_end_mask_0 = const()[name = tensor<string, []>("op_12457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12457_cast = slice_by_index(begin = var_12457_begin_0, end = var_12457_end_0, end_mask = var_12457_end_mask_0, x = var_12425_cast)[name = tensor<string, []>("op_12457_cast")];
            tensor<int32, [4]> var_12458_begin_0 = const()[name = tensor<string, []>("op_12458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12458_end_0 = const()[name = tensor<string, []>("op_12458_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12458_end_mask_0 = const()[name = tensor<string, []>("op_12458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12458_cast = slice_by_index(begin = var_12458_begin_0, end = var_12458_end_0, end_mask = var_12458_end_mask_0, x = var_12429_cast)[name = tensor<string, []>("op_12458_cast")];
            tensor<int32, [4]> var_12459_begin_0 = const()[name = tensor<string, []>("op_12459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12459_end_0 = const()[name = tensor<string, []>("op_12459_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12459_end_mask_0 = const()[name = tensor<string, []>("op_12459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12459_cast = slice_by_index(begin = var_12459_begin_0, end = var_12459_end_0, end_mask = var_12459_end_mask_0, x = var_12429_cast)[name = tensor<string, []>("op_12459_cast")];
            tensor<int32, [4]> var_12460_begin_0 = const()[name = tensor<string, []>("op_12460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12460_end_0 = const()[name = tensor<string, []>("op_12460_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12460_end_mask_0 = const()[name = tensor<string, []>("op_12460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12460_cast = slice_by_index(begin = var_12460_begin_0, end = var_12460_end_0, end_mask = var_12460_end_mask_0, x = var_12433_cast)[name = tensor<string, []>("op_12460_cast")];
            tensor<int32, [4]> var_12461_begin_0 = const()[name = tensor<string, []>("op_12461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12461_end_0 = const()[name = tensor<string, []>("op_12461_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12461_end_mask_0 = const()[name = tensor<string, []>("op_12461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12461_cast = slice_by_index(begin = var_12461_begin_0, end = var_12461_end_0, end_mask = var_12461_end_mask_0, x = var_12433_cast)[name = tensor<string, []>("op_12461_cast")];
            tensor<int32, [4]> var_12462_begin_0 = const()[name = tensor<string, []>("op_12462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12462_end_0 = const()[name = tensor<string, []>("op_12462_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12462_end_mask_0 = const()[name = tensor<string, []>("op_12462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12462_cast = slice_by_index(begin = var_12462_begin_0, end = var_12462_end_0, end_mask = var_12462_end_mask_0, x = var_12437_cast)[name = tensor<string, []>("op_12462_cast")];
            tensor<int32, [4]> var_12463_begin_0 = const()[name = tensor<string, []>("op_12463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12463_end_0 = const()[name = tensor<string, []>("op_12463_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12463_end_mask_0 = const()[name = tensor<string, []>("op_12463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12463_cast = slice_by_index(begin = var_12463_begin_0, end = var_12463_end_0, end_mask = var_12463_end_mask_0, x = var_12437_cast)[name = tensor<string, []>("op_12463_cast")];
            tensor<int32, [4]> var_12464_begin_0 = const()[name = tensor<string, []>("op_12464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12464_end_0 = const()[name = tensor<string, []>("op_12464_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12464_end_mask_0 = const()[name = tensor<string, []>("op_12464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12464_cast = slice_by_index(begin = var_12464_begin_0, end = var_12464_end_0, end_mask = var_12464_end_mask_0, x = var_12441_cast)[name = tensor<string, []>("op_12464_cast")];
            tensor<int32, [4]> var_12465_begin_0 = const()[name = tensor<string, []>("op_12465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12465_end_0 = const()[name = tensor<string, []>("op_12465_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12465_end_mask_0 = const()[name = tensor<string, []>("op_12465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12465_cast = slice_by_index(begin = var_12465_begin_0, end = var_12465_end_0, end_mask = var_12465_end_mask_0, x = var_12441_cast)[name = tensor<string, []>("op_12465_cast")];
            tensor<int32, [4]> var_12466_begin_0 = const()[name = tensor<string, []>("op_12466_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12466_end_0 = const()[name = tensor<string, []>("op_12466_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12466_end_mask_0 = const()[name = tensor<string, []>("op_12466_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12466_cast = slice_by_index(begin = var_12466_begin_0, end = var_12466_end_0, end_mask = var_12466_end_mask_0, x = var_12445_cast)[name = tensor<string, []>("op_12466_cast")];
            tensor<int32, [4]> var_12467_begin_0 = const()[name = tensor<string, []>("op_12467_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12467_end_0 = const()[name = tensor<string, []>("op_12467_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12467_end_mask_0 = const()[name = tensor<string, []>("op_12467_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12467_cast = slice_by_index(begin = var_12467_begin_0, end = var_12467_end_0, end_mask = var_12467_end_mask_0, x = var_12445_cast)[name = tensor<string, []>("op_12467_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_12472_begin_0 = const()[name = tensor<string, []>("op_12472_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12472_end_0 = const()[name = tensor<string, []>("op_12472_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_12472_end_mask_0 = const()[name = tensor<string, []>("op_12472_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, 64]> var_12472_cast = slice_by_index(begin = var_12472_begin_0, end = var_12472_end_0, end_mask = var_12472_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12472_cast")];
            tensor<int32, [4]> var_12476_begin_0 = const()[name = tensor<string, []>("op_12476_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_12476_end_0 = const()[name = tensor<string, []>("op_12476_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_12476_end_mask_0 = const()[name = tensor<string, []>("op_12476_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12476_cast = slice_by_index(begin = var_12476_begin_0, end = var_12476_end_0, end_mask = var_12476_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12476_cast")];
            tensor<int32, [4]> var_12480_begin_0 = const()[name = tensor<string, []>("op_12480_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_12480_end_0 = const()[name = tensor<string, []>("op_12480_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_12480_end_mask_0 = const()[name = tensor<string, []>("op_12480_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12480_cast = slice_by_index(begin = var_12480_begin_0, end = var_12480_end_0, end_mask = var_12480_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12480_cast")];
            tensor<int32, [4]> var_12484_begin_0 = const()[name = tensor<string, []>("op_12484_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_12484_end_0 = const()[name = tensor<string, []>("op_12484_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_12484_end_mask_0 = const()[name = tensor<string, []>("op_12484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12484_cast = slice_by_index(begin = var_12484_begin_0, end = var_12484_end_0, end_mask = var_12484_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12484_cast")];
            tensor<int32, [4]> var_12488_begin_0 = const()[name = tensor<string, []>("op_12488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_12488_end_0 = const()[name = tensor<string, []>("op_12488_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_12488_end_mask_0 = const()[name = tensor<string, []>("op_12488_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12488_cast = slice_by_index(begin = var_12488_begin_0, end = var_12488_end_0, end_mask = var_12488_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12488_cast")];
            tensor<int32, [4]> var_12492_begin_0 = const()[name = tensor<string, []>("op_12492_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_12492_end_0 = const()[name = tensor<string, []>("op_12492_end_0"), val = tensor<int32, [4]>([2, 77, 1, 384])];
            tensor<bool, [4]> var_12492_end_mask_0 = const()[name = tensor<string, []>("op_12492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12492_cast = slice_by_index(begin = var_12492_begin_0, end = var_12492_end_0, end_mask = var_12492_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12492_cast")];
            tensor<int32, [4]> var_12496_begin_0 = const()[name = tensor<string, []>("op_12496_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 384])];
            tensor<int32, [4]> var_12496_end_0 = const()[name = tensor<string, []>("op_12496_end_0"), val = tensor<int32, [4]>([2, 77, 1, 448])];
            tensor<bool, [4]> var_12496_end_mask_0 = const()[name = tensor<string, []>("op_12496_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12496_cast = slice_by_index(begin = var_12496_begin_0, end = var_12496_end_0, end_mask = var_12496_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12496_cast")];
            tensor<int32, [4]> var_12500_begin_0 = const()[name = tensor<string, []>("op_12500_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 448])];
            tensor<int32, [4]> var_12500_end_0 = const()[name = tensor<string, []>("op_12500_end_0"), val = tensor<int32, [4]>([2, 77, 1, 512])];
            tensor<bool, [4]> var_12500_end_mask_0 = const()[name = tensor<string, []>("op_12500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12500_cast = slice_by_index(begin = var_12500_begin_0, end = var_12500_end_0, end_mask = var_12500_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12500_cast")];
            tensor<int32, [4]> var_12504_begin_0 = const()[name = tensor<string, []>("op_12504_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12504_end_0 = const()[name = tensor<string, []>("op_12504_end_0"), val = tensor<int32, [4]>([2, 77, 1, 576])];
            tensor<bool, [4]> var_12504_end_mask_0 = const()[name = tensor<string, []>("op_12504_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12504_cast = slice_by_index(begin = var_12504_begin_0, end = var_12504_end_0, end_mask = var_12504_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12504_cast")];
            tensor<int32, [4]> var_12508_begin_0 = const()[name = tensor<string, []>("op_12508_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 576])];
            tensor<int32, [4]> var_12508_end_0 = const()[name = tensor<string, []>("op_12508_end_0"), val = tensor<int32, [4]>([2, 77, 1, 640])];
            tensor<bool, [4]> var_12508_end_mask_0 = const()[name = tensor<string, []>("op_12508_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_12508_cast = slice_by_index(begin = var_12508_begin_0, end = var_12508_end_0, end_mask = var_12508_end_mask_0, x = transpose_6)[name = tensor<string, []>("op_12508_cast")];
            tensor<int32, [4]> var_12510_begin_0 = const()[name = tensor<string, []>("op_12510_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12510_end_0 = const()[name = tensor<string, []>("op_12510_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_12510_end_mask_0 = const()[name = tensor<string, []>("op_12510_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12510_cast = slice_by_index(begin = var_12510_begin_0, end = var_12510_end_0, end_mask = var_12510_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12510_cast")];
            tensor<int32, [4]> var_12514_begin_0 = const()[name = tensor<string, []>("op_12514_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12514_end_0 = const()[name = tensor<string, []>("op_12514_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_12514_end_mask_0 = const()[name = tensor<string, []>("op_12514_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12514_cast = slice_by_index(begin = var_12514_begin_0, end = var_12514_end_0, end_mask = var_12514_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12514_cast")];
            tensor<int32, [4]> var_12518_begin_0 = const()[name = tensor<string, []>("op_12518_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12518_end_0 = const()[name = tensor<string, []>("op_12518_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_12518_end_mask_0 = const()[name = tensor<string, []>("op_12518_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12518_cast = slice_by_index(begin = var_12518_begin_0, end = var_12518_end_0, end_mask = var_12518_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12518_cast")];
            tensor<int32, [4]> var_12522_begin_0 = const()[name = tensor<string, []>("op_12522_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12522_end_0 = const()[name = tensor<string, []>("op_12522_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_12522_end_mask_0 = const()[name = tensor<string, []>("op_12522_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12522_cast = slice_by_index(begin = var_12522_begin_0, end = var_12522_end_0, end_mask = var_12522_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12522_cast")];
            tensor<int32, [4]> var_12526_begin_0 = const()[name = tensor<string, []>("op_12526_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12526_end_0 = const()[name = tensor<string, []>("op_12526_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_12526_end_mask_0 = const()[name = tensor<string, []>("op_12526_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12526_cast = slice_by_index(begin = var_12526_begin_0, end = var_12526_end_0, end_mask = var_12526_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12526_cast")];
            tensor<int32, [4]> var_12530_begin_0 = const()[name = tensor<string, []>("op_12530_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_12530_end_0 = const()[name = tensor<string, []>("op_12530_end_0"), val = tensor<int32, [4]>([2, 384, 1, 77])];
            tensor<bool, [4]> var_12530_end_mask_0 = const()[name = tensor<string, []>("op_12530_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12530_cast = slice_by_index(begin = var_12530_begin_0, end = var_12530_end_0, end_mask = var_12530_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12530_cast")];
            tensor<int32, [4]> var_12534_begin_0 = const()[name = tensor<string, []>("op_12534_begin_0"), val = tensor<int32, [4]>([0, 384, 0, 0])];
            tensor<int32, [4]> var_12534_end_0 = const()[name = tensor<string, []>("op_12534_end_0"), val = tensor<int32, [4]>([2, 448, 1, 77])];
            tensor<bool, [4]> var_12534_end_mask_0 = const()[name = tensor<string, []>("op_12534_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12534_cast = slice_by_index(begin = var_12534_begin_0, end = var_12534_end_0, end_mask = var_12534_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12534_cast")];
            tensor<int32, [4]> var_12538_begin_0 = const()[name = tensor<string, []>("op_12538_begin_0"), val = tensor<int32, [4]>([0, 448, 0, 0])];
            tensor<int32, [4]> var_12538_end_0 = const()[name = tensor<string, []>("op_12538_end_0"), val = tensor<int32, [4]>([2, 512, 1, 77])];
            tensor<bool, [4]> var_12538_end_mask_0 = const()[name = tensor<string, []>("op_12538_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12538_cast = slice_by_index(begin = var_12538_begin_0, end = var_12538_end_0, end_mask = var_12538_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12538_cast")];
            tensor<int32, [4]> var_12542_begin_0 = const()[name = tensor<string, []>("op_12542_begin_0"), val = tensor<int32, [4]>([0, 512, 0, 0])];
            tensor<int32, [4]> var_12542_end_0 = const()[name = tensor<string, []>("op_12542_end_0"), val = tensor<int32, [4]>([2, 576, 1, 77])];
            tensor<bool, [4]> var_12542_end_mask_0 = const()[name = tensor<string, []>("op_12542_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12542_cast = slice_by_index(begin = var_12542_begin_0, end = var_12542_end_0, end_mask = var_12542_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12542_cast")];
            tensor<int32, [4]> var_12546_begin_0 = const()[name = tensor<string, []>("op_12546_begin_0"), val = tensor<int32, [4]>([0, 576, 0, 0])];
            tensor<int32, [4]> var_12546_end_0 = const()[name = tensor<string, []>("op_12546_end_0"), val = tensor<int32, [4]>([2, 640, 1, 77])];
            tensor<bool, [4]> var_12546_end_mask_0 = const()[name = tensor<string, []>("op_12546_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_12546_cast = slice_by_index(begin = var_12546_begin_0, end = var_12546_end_0, end_mask = var_12546_end_mask_0, x = v_51_cast)[name = tensor<string, []>("op_12546_cast")];
            tensor<string, []> var_12550_equation_0 = const()[name = tensor<string, []>("op_12550_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12550_cast = einsum(equation = var_12550_equation_0, values = (var_12472_cast, var_12448_cast))[name = tensor<string, []>("op_12550_cast")];
            tensor<fp16, []> var_12551_to_fp16 = const()[name = tensor<string, []>("op_12551_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_681_cast = mul(x = var_12550_cast, y = var_12551_to_fp16)[name = tensor<string, []>("aw_chunk_681_cast")];
            tensor<string, []> var_12554_equation_0 = const()[name = tensor<string, []>("op_12554_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12554_cast = einsum(equation = var_12554_equation_0, values = (var_12472_cast, var_12449_cast))[name = tensor<string, []>("op_12554_cast")];
            tensor<fp16, []> var_12555_to_fp16 = const()[name = tensor<string, []>("op_12555_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_683_cast = mul(x = var_12554_cast, y = var_12555_to_fp16)[name = tensor<string, []>("aw_chunk_683_cast")];
            tensor<string, []> var_12558_equation_0 = const()[name = tensor<string, []>("op_12558_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12558_cast = einsum(equation = var_12558_equation_0, values = (var_12476_cast, var_12450_cast))[name = tensor<string, []>("op_12558_cast")];
            tensor<fp16, []> var_12559_to_fp16 = const()[name = tensor<string, []>("op_12559_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_685_cast = mul(x = var_12558_cast, y = var_12559_to_fp16)[name = tensor<string, []>("aw_chunk_685_cast")];
            tensor<string, []> var_12562_equation_0 = const()[name = tensor<string, []>("op_12562_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12562_cast = einsum(equation = var_12562_equation_0, values = (var_12476_cast, var_12451_cast))[name = tensor<string, []>("op_12562_cast")];
            tensor<fp16, []> var_12563_to_fp16 = const()[name = tensor<string, []>("op_12563_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_687_cast = mul(x = var_12562_cast, y = var_12563_to_fp16)[name = tensor<string, []>("aw_chunk_687_cast")];
            tensor<string, []> var_12566_equation_0 = const()[name = tensor<string, []>("op_12566_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12566_cast = einsum(equation = var_12566_equation_0, values = (var_12480_cast, var_12452_cast))[name = tensor<string, []>("op_12566_cast")];
            tensor<fp16, []> var_12567_to_fp16 = const()[name = tensor<string, []>("op_12567_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_689_cast = mul(x = var_12566_cast, y = var_12567_to_fp16)[name = tensor<string, []>("aw_chunk_689_cast")];
            tensor<string, []> var_12570_equation_0 = const()[name = tensor<string, []>("op_12570_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12570_cast = einsum(equation = var_12570_equation_0, values = (var_12480_cast, var_12453_cast))[name = tensor<string, []>("op_12570_cast")];
            tensor<fp16, []> var_12571_to_fp16 = const()[name = tensor<string, []>("op_12571_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_691_cast = mul(x = var_12570_cast, y = var_12571_to_fp16)[name = tensor<string, []>("aw_chunk_691_cast")];
            tensor<string, []> var_12574_equation_0 = const()[name = tensor<string, []>("op_12574_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12574_cast = einsum(equation = var_12574_equation_0, values = (var_12484_cast, var_12454_cast))[name = tensor<string, []>("op_12574_cast")];
            tensor<fp16, []> var_12575_to_fp16 = const()[name = tensor<string, []>("op_12575_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_693_cast = mul(x = var_12574_cast, y = var_12575_to_fp16)[name = tensor<string, []>("aw_chunk_693_cast")];
            tensor<string, []> var_12578_equation_0 = const()[name = tensor<string, []>("op_12578_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12578_cast = einsum(equation = var_12578_equation_0, values = (var_12484_cast, var_12455_cast))[name = tensor<string, []>("op_12578_cast")];
            tensor<fp16, []> var_12579_to_fp16 = const()[name = tensor<string, []>("op_12579_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_695_cast = mul(x = var_12578_cast, y = var_12579_to_fp16)[name = tensor<string, []>("aw_chunk_695_cast")];
            tensor<string, []> var_12582_equation_0 = const()[name = tensor<string, []>("op_12582_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12582_cast = einsum(equation = var_12582_equation_0, values = (var_12488_cast, var_12456_cast))[name = tensor<string, []>("op_12582_cast")];
            tensor<fp16, []> var_12583_to_fp16 = const()[name = tensor<string, []>("op_12583_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_697_cast = mul(x = var_12582_cast, y = var_12583_to_fp16)[name = tensor<string, []>("aw_chunk_697_cast")];
            tensor<string, []> var_12586_equation_0 = const()[name = tensor<string, []>("op_12586_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12586_cast = einsum(equation = var_12586_equation_0, values = (var_12488_cast, var_12457_cast))[name = tensor<string, []>("op_12586_cast")];
            tensor<fp16, []> var_12587_to_fp16 = const()[name = tensor<string, []>("op_12587_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_699_cast = mul(x = var_12586_cast, y = var_12587_to_fp16)[name = tensor<string, []>("aw_chunk_699_cast")];
            tensor<string, []> var_12590_equation_0 = const()[name = tensor<string, []>("op_12590_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12590_cast = einsum(equation = var_12590_equation_0, values = (var_12492_cast, var_12458_cast))[name = tensor<string, []>("op_12590_cast")];
            tensor<fp16, []> var_12591_to_fp16 = const()[name = tensor<string, []>("op_12591_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_701_cast = mul(x = var_12590_cast, y = var_12591_to_fp16)[name = tensor<string, []>("aw_chunk_701_cast")];
            tensor<string, []> var_12594_equation_0 = const()[name = tensor<string, []>("op_12594_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12594_cast = einsum(equation = var_12594_equation_0, values = (var_12492_cast, var_12459_cast))[name = tensor<string, []>("op_12594_cast")];
            tensor<fp16, []> var_12595_to_fp16 = const()[name = tensor<string, []>("op_12595_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_703_cast = mul(x = var_12594_cast, y = var_12595_to_fp16)[name = tensor<string, []>("aw_chunk_703_cast")];
            tensor<string, []> var_12598_equation_0 = const()[name = tensor<string, []>("op_12598_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12598_cast = einsum(equation = var_12598_equation_0, values = (var_12496_cast, var_12460_cast))[name = tensor<string, []>("op_12598_cast")];
            tensor<fp16, []> var_12599_to_fp16 = const()[name = tensor<string, []>("op_12599_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_705_cast = mul(x = var_12598_cast, y = var_12599_to_fp16)[name = tensor<string, []>("aw_chunk_705_cast")];
            tensor<string, []> var_12602_equation_0 = const()[name = tensor<string, []>("op_12602_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12602_cast = einsum(equation = var_12602_equation_0, values = (var_12496_cast, var_12461_cast))[name = tensor<string, []>("op_12602_cast")];
            tensor<fp16, []> var_12603_to_fp16 = const()[name = tensor<string, []>("op_12603_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_707_cast = mul(x = var_12602_cast, y = var_12603_to_fp16)[name = tensor<string, []>("aw_chunk_707_cast")];
            tensor<string, []> var_12606_equation_0 = const()[name = tensor<string, []>("op_12606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12606_cast = einsum(equation = var_12606_equation_0, values = (var_12500_cast, var_12462_cast))[name = tensor<string, []>("op_12606_cast")];
            tensor<fp16, []> var_12607_to_fp16 = const()[name = tensor<string, []>("op_12607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_709_cast = mul(x = var_12606_cast, y = var_12607_to_fp16)[name = tensor<string, []>("aw_chunk_709_cast")];
            tensor<string, []> var_12610_equation_0 = const()[name = tensor<string, []>("op_12610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12610_cast = einsum(equation = var_12610_equation_0, values = (var_12500_cast, var_12463_cast))[name = tensor<string, []>("op_12610_cast")];
            tensor<fp16, []> var_12611_to_fp16 = const()[name = tensor<string, []>("op_12611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_711_cast = mul(x = var_12610_cast, y = var_12611_to_fp16)[name = tensor<string, []>("aw_chunk_711_cast")];
            tensor<string, []> var_12614_equation_0 = const()[name = tensor<string, []>("op_12614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12614_cast = einsum(equation = var_12614_equation_0, values = (var_12504_cast, var_12464_cast))[name = tensor<string, []>("op_12614_cast")];
            tensor<fp16, []> var_12615_to_fp16 = const()[name = tensor<string, []>("op_12615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_713_cast = mul(x = var_12614_cast, y = var_12615_to_fp16)[name = tensor<string, []>("aw_chunk_713_cast")];
            tensor<string, []> var_12618_equation_0 = const()[name = tensor<string, []>("op_12618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12618_cast = einsum(equation = var_12618_equation_0, values = (var_12504_cast, var_12465_cast))[name = tensor<string, []>("op_12618_cast")];
            tensor<fp16, []> var_12619_to_fp16 = const()[name = tensor<string, []>("op_12619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_715_cast = mul(x = var_12618_cast, y = var_12619_to_fp16)[name = tensor<string, []>("aw_chunk_715_cast")];
            tensor<string, []> var_12622_equation_0 = const()[name = tensor<string, []>("op_12622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12622_cast = einsum(equation = var_12622_equation_0, values = (var_12508_cast, var_12466_cast))[name = tensor<string, []>("op_12622_cast")];
            tensor<fp16, []> var_12623_to_fp16 = const()[name = tensor<string, []>("op_12623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_717_cast = mul(x = var_12622_cast, y = var_12623_to_fp16)[name = tensor<string, []>("aw_chunk_717_cast")];
            tensor<string, []> var_12626_equation_0 = const()[name = tensor<string, []>("op_12626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_12626_cast = einsum(equation = var_12626_equation_0, values = (var_12508_cast, var_12467_cast))[name = tensor<string, []>("op_12626_cast")];
            tensor<fp16, []> var_12627_to_fp16 = const()[name = tensor<string, []>("op_12627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_719_cast = mul(x = var_12626_cast, y = var_12627_to_fp16)[name = tensor<string, []>("aw_chunk_719_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12629_cast = softmax(axis = var_10257, x = aw_chunk_681_cast)[name = tensor<string, []>("op_12629_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12630_cast = softmax(axis = var_10257, x = aw_chunk_683_cast)[name = tensor<string, []>("op_12630_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12631_cast = softmax(axis = var_10257, x = aw_chunk_685_cast)[name = tensor<string, []>("op_12631_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12632_cast = softmax(axis = var_10257, x = aw_chunk_687_cast)[name = tensor<string, []>("op_12632_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12633_cast = softmax(axis = var_10257, x = aw_chunk_689_cast)[name = tensor<string, []>("op_12633_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12634_cast = softmax(axis = var_10257, x = aw_chunk_691_cast)[name = tensor<string, []>("op_12634_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12635_cast = softmax(axis = var_10257, x = aw_chunk_693_cast)[name = tensor<string, []>("op_12635_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12636_cast = softmax(axis = var_10257, x = aw_chunk_695_cast)[name = tensor<string, []>("op_12636_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12637_cast = softmax(axis = var_10257, x = aw_chunk_697_cast)[name = tensor<string, []>("op_12637_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12638_cast = softmax(axis = var_10257, x = aw_chunk_699_cast)[name = tensor<string, []>("op_12638_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12639_cast = softmax(axis = var_10257, x = aw_chunk_701_cast)[name = tensor<string, []>("op_12639_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12640_cast = softmax(axis = var_10257, x = aw_chunk_703_cast)[name = tensor<string, []>("op_12640_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12641_cast = softmax(axis = var_10257, x = aw_chunk_705_cast)[name = tensor<string, []>("op_12641_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12642_cast = softmax(axis = var_10257, x = aw_chunk_707_cast)[name = tensor<string, []>("op_12642_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12643_cast = softmax(axis = var_10257, x = aw_chunk_709_cast)[name = tensor<string, []>("op_12643_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12644_cast = softmax(axis = var_10257, x = aw_chunk_711_cast)[name = tensor<string, []>("op_12644_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12645_cast = softmax(axis = var_10257, x = aw_chunk_713_cast)[name = tensor<string, []>("op_12645_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12646_cast = softmax(axis = var_10257, x = aw_chunk_715_cast)[name = tensor<string, []>("op_12646_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12647_cast = softmax(axis = var_10257, x = aw_chunk_717_cast)[name = tensor<string, []>("op_12647_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_12648_cast = softmax(axis = var_10257, x = aw_chunk_719_cast)[name = tensor<string, []>("op_12648_cast")];
            tensor<string, []> var_12650_equation_0 = const()[name = tensor<string, []>("op_12650_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12650_cast = einsum(equation = var_12650_equation_0, values = (var_12510_cast, var_12629_cast))[name = tensor<string, []>("op_12650_cast")];
            tensor<string, []> var_12652_equation_0 = const()[name = tensor<string, []>("op_12652_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12652_cast = einsum(equation = var_12652_equation_0, values = (var_12510_cast, var_12630_cast))[name = tensor<string, []>("op_12652_cast")];
            tensor<string, []> var_12654_equation_0 = const()[name = tensor<string, []>("op_12654_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12654_cast = einsum(equation = var_12654_equation_0, values = (var_12514_cast, var_12631_cast))[name = tensor<string, []>("op_12654_cast")];
            tensor<string, []> var_12656_equation_0 = const()[name = tensor<string, []>("op_12656_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12656_cast = einsum(equation = var_12656_equation_0, values = (var_12514_cast, var_12632_cast))[name = tensor<string, []>("op_12656_cast")];
            tensor<string, []> var_12658_equation_0 = const()[name = tensor<string, []>("op_12658_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12658_cast = einsum(equation = var_12658_equation_0, values = (var_12518_cast, var_12633_cast))[name = tensor<string, []>("op_12658_cast")];
            tensor<string, []> var_12660_equation_0 = const()[name = tensor<string, []>("op_12660_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12660_cast = einsum(equation = var_12660_equation_0, values = (var_12518_cast, var_12634_cast))[name = tensor<string, []>("op_12660_cast")];
            tensor<string, []> var_12662_equation_0 = const()[name = tensor<string, []>("op_12662_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12662_cast = einsum(equation = var_12662_equation_0, values = (var_12522_cast, var_12635_cast))[name = tensor<string, []>("op_12662_cast")];
            tensor<string, []> var_12664_equation_0 = const()[name = tensor<string, []>("op_12664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12664_cast = einsum(equation = var_12664_equation_0, values = (var_12522_cast, var_12636_cast))[name = tensor<string, []>("op_12664_cast")];
            tensor<string, []> var_12666_equation_0 = const()[name = tensor<string, []>("op_12666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12666_cast = einsum(equation = var_12666_equation_0, values = (var_12526_cast, var_12637_cast))[name = tensor<string, []>("op_12666_cast")];
            tensor<string, []> var_12668_equation_0 = const()[name = tensor<string, []>("op_12668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12668_cast = einsum(equation = var_12668_equation_0, values = (var_12526_cast, var_12638_cast))[name = tensor<string, []>("op_12668_cast")];
            tensor<string, []> var_12670_equation_0 = const()[name = tensor<string, []>("op_12670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12670_cast = einsum(equation = var_12670_equation_0, values = (var_12530_cast, var_12639_cast))[name = tensor<string, []>("op_12670_cast")];
            tensor<string, []> var_12672_equation_0 = const()[name = tensor<string, []>("op_12672_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12672_cast = einsum(equation = var_12672_equation_0, values = (var_12530_cast, var_12640_cast))[name = tensor<string, []>("op_12672_cast")];
            tensor<string, []> var_12674_equation_0 = const()[name = tensor<string, []>("op_12674_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12674_cast = einsum(equation = var_12674_equation_0, values = (var_12534_cast, var_12641_cast))[name = tensor<string, []>("op_12674_cast")];
            tensor<string, []> var_12676_equation_0 = const()[name = tensor<string, []>("op_12676_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12676_cast = einsum(equation = var_12676_equation_0, values = (var_12534_cast, var_12642_cast))[name = tensor<string, []>("op_12676_cast")];
            tensor<string, []> var_12678_equation_0 = const()[name = tensor<string, []>("op_12678_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12678_cast = einsum(equation = var_12678_equation_0, values = (var_12538_cast, var_12643_cast))[name = tensor<string, []>("op_12678_cast")];
            tensor<string, []> var_12680_equation_0 = const()[name = tensor<string, []>("op_12680_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12680_cast = einsum(equation = var_12680_equation_0, values = (var_12538_cast, var_12644_cast))[name = tensor<string, []>("op_12680_cast")];
            tensor<string, []> var_12682_equation_0 = const()[name = tensor<string, []>("op_12682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12682_cast = einsum(equation = var_12682_equation_0, values = (var_12542_cast, var_12645_cast))[name = tensor<string, []>("op_12682_cast")];
            tensor<string, []> var_12684_equation_0 = const()[name = tensor<string, []>("op_12684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12684_cast = einsum(equation = var_12684_equation_0, values = (var_12542_cast, var_12646_cast))[name = tensor<string, []>("op_12684_cast")];
            tensor<string, []> var_12686_equation_0 = const()[name = tensor<string, []>("op_12686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12686_cast = einsum(equation = var_12686_equation_0, values = (var_12546_cast, var_12647_cast))[name = tensor<string, []>("op_12686_cast")];
            tensor<string, []> var_12688_equation_0 = const()[name = tensor<string, []>("op_12688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_12688_cast = einsum(equation = var_12688_equation_0, values = (var_12546_cast, var_12648_cast))[name = tensor<string, []>("op_12688_cast")];
            tensor<bool, []> var_12690_interleave_0 = const()[name = tensor<string, []>("op_12690_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12690_cast = concat(axis = var_10234, interleave = var_12690_interleave_0, values = (var_12650_cast, var_12652_cast))[name = tensor<string, []>("op_12690_cast")];
            tensor<bool, []> var_12692_interleave_0 = const()[name = tensor<string, []>("op_12692_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12692_cast = concat(axis = var_10234, interleave = var_12692_interleave_0, values = (var_12654_cast, var_12656_cast))[name = tensor<string, []>("op_12692_cast")];
            tensor<bool, []> var_12694_interleave_0 = const()[name = tensor<string, []>("op_12694_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12694_cast = concat(axis = var_10234, interleave = var_12694_interleave_0, values = (var_12658_cast, var_12660_cast))[name = tensor<string, []>("op_12694_cast")];
            tensor<bool, []> var_12696_interleave_0 = const()[name = tensor<string, []>("op_12696_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12696_cast = concat(axis = var_10234, interleave = var_12696_interleave_0, values = (var_12662_cast, var_12664_cast))[name = tensor<string, []>("op_12696_cast")];
            tensor<bool, []> var_12698_interleave_0 = const()[name = tensor<string, []>("op_12698_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12698_cast = concat(axis = var_10234, interleave = var_12698_interleave_0, values = (var_12666_cast, var_12668_cast))[name = tensor<string, []>("op_12698_cast")];
            tensor<bool, []> var_12700_interleave_0 = const()[name = tensor<string, []>("op_12700_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12700_cast = concat(axis = var_10234, interleave = var_12700_interleave_0, values = (var_12670_cast, var_12672_cast))[name = tensor<string, []>("op_12700_cast")];
            tensor<bool, []> var_12702_interleave_0 = const()[name = tensor<string, []>("op_12702_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12702_cast = concat(axis = var_10234, interleave = var_12702_interleave_0, values = (var_12674_cast, var_12676_cast))[name = tensor<string, []>("op_12702_cast")];
            tensor<bool, []> var_12704_interleave_0 = const()[name = tensor<string, []>("op_12704_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12704_cast = concat(axis = var_10234, interleave = var_12704_interleave_0, values = (var_12678_cast, var_12680_cast))[name = tensor<string, []>("op_12704_cast")];
            tensor<bool, []> var_12706_interleave_0 = const()[name = tensor<string, []>("op_12706_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12706_cast = concat(axis = var_10234, interleave = var_12706_interleave_0, values = (var_12682_cast, var_12684_cast))[name = tensor<string, []>("op_12706_cast")];
            tensor<bool, []> var_12708_interleave_0 = const()[name = tensor<string, []>("op_12708_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 1024]> var_12708_cast = concat(axis = var_10234, interleave = var_12708_interleave_0, values = (var_12686_cast, var_12688_cast))[name = tensor<string, []>("op_12708_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_10257, interleave = input_437_interleave_0, values = (var_12690_cast, var_12692_cast, var_12694_cast, var_12696_cast, var_12698_cast, var_12700_cast, var_12702_cast, var_12704_cast, var_12706_cast, var_12708_cast))[name = tensor<string, []>("input_437_cast")];
            tensor<int32, [2]> var_12714 = const()[name = tensor<string, []>("op_12714"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12716 = const()[name = tensor<string, []>("op_12716"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12718_pad_type_0 = const()[name = tensor<string, []>("op_12718_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12718_pad_0 = const()[name = tensor<string, []>("op_12718_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209859840))), lut = tensor<fp16, [4]>([-0x1.5a8p-6, -0x1.874p-8, 0x1.924p-8, 0x1.608p-6]), 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, []>(209962304)))];
            tensor<fp16, [2, 640, 1, 1024]> var_12718_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_12716, groups = var_10257, pad = var_12718_pad_0, pad_type = var_12718_pad_type_0, strides = var_12714, 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_12718_cast")];
            tensor<fp16, [2, 640, 1, 1024]> inputs_77_cast = add(x = var_12718_cast, y = inputs_75_cast)[name = tensor<string, []>("inputs_77_cast")];
            tensor<int32, [1]> var_12722 = const()[name = tensor<string, []>("op_12722"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> channels_mean_77_cast = reduce_mean(axes = var_12722, keep_dims = var_10252, 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_12726 = const()[name = tensor<string, []>("op_12726"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 1024]> var_12727_cast = reduce_mean(axes = var_12726, keep_dims = var_10252, x = zero_mean_sq_77_cast)[name = tensor<string, []>("op_12727_cast")];
            tensor<fp16, []> var_12728_to_fp16 = const()[name = tensor<string, []>("op_12728_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 1024]> var_12729_cast = add(x = var_12727_cast, y = var_12728_to_fp16)[name = tensor<string, []>("op_12729_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_12729_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_12733_to_fp16 = const()[name = tensor<string, []>("op_12733_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209963648)))];
            tensor<fp16, [2, 640, 1, 1024]> var_12734_cast = add(x = out_77_cast, y = var_12733_to_fp16)[name = tensor<string, []>("op_12734_cast")];
            tensor<fp16, [1, 640, 1, 1]> var_12736_to_fp16 = const()[name = tensor<string, []>("op_12736_to_fp16"), val = tensor<fp16, [1, 640, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209964992)))];
            tensor<fp16, [2, 640, 1, 1024]> input_439_cast = mul(x = var_12734_cast, y = var_12736_to_fp16)[name = tensor<string, []>("input_439_cast")];
            tensor<int32, [2]> var_12744 = const()[name = tensor<string, []>("op_12744"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12746 = const()[name = tensor<string, []>("op_12746"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12748_pad_type_0 = const()[name = tensor<string, []>("op_12748_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12748_pad_0 = const()[name = tensor<string, []>("op_12748_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, [819200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209966336))), lut = tensor<fp16, [4]>([-0x1.8bp-4, -0x1.cbcp-6, 0x1.c6cp-6, 0x1.894p-4]), 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, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210785600))), lut = tensor<fp16, [4]>([-0x1.72cp-4, 0x1.578p-5, -0x1.254p-2, -0x1.a8cp-7]), 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_12748_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_12746, groups = var_10257, pad = var_12748_pad_0, pad_type = var_12748_pad_type_0, strides = var_12744, 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_12748_cast")];
            tensor<int32, [2]> var_12749_split_sizes_0 = const()[name = tensor<string, []>("op_12749_split_sizes_0"), val = tensor<int32, [2]>([2560, 2560])];
            tensor<int32, []> var_12749_axis_0 = const()[name = tensor<string, []>("op_12749_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 2560, 1, 1024]> var_12749_cast_0, tensor<fp16, [2, 2560, 1, 1024]> var_12749_cast_1 = split(axis = var_12749_axis_0, split_sizes = var_12749_split_sizes_0, x = var_12748_cast)[name = tensor<string, []>("op_12749_cast")];
            tensor<string, []> var_12751_mode_0 = const()[name = tensor<string, []>("op_12751_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 2560, 1, 1024]> var_12751_cast = gelu(mode = var_12751_mode_0, x = var_12749_cast_1)[name = tensor<string, []>("op_12751_cast")];
            tensor<fp16, [2, 2560, 1, 1024]> input_441_cast = mul(x = var_12749_cast_0, y = var_12751_cast)[name = tensor<string, []>("input_441_cast")];
            tensor<int32, [2]> var_12755 = const()[name = tensor<string, []>("op_12755"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12757 = const()[name = tensor<string, []>("op_12757"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_12759_pad_type_0 = const()[name = tensor<string, []>("op_12759_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_12759_pad_0 = const()[name = tensor<string, []>("op_12759_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, [409600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210786944))), lut = tensor<fp16, [4]>([-0x1.8bp-4, -0x1.d34p-6, 0x1.d44p-6, 0x1.8b4p-4]), 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, []>(211196608)))];
            tensor<fp16, [2, 640, 1, 1024]> var_12759_cast = conv(bias = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_12757, groups = var_10257, pad = var_12759_pad_0, pad_type = var_12759_pad_type_0, strides = var_12755, weight = up_blocks_2_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_441_cast)[name = tensor<string, []>("op_12759_cast")];
            tensor<fp16, [2, 640, 1, 1024]> hidden_states_273_cast = add(x = var_12759_cast, y = inputs_77_cast)[name = tensor<string, []>("hidden_states_273_cast")];
            tensor<int32, [4]> var_12761 = const()[name = tensor<string, []>("op_12761"), val = tensor<int32, [4]>([2, 640, 32, 32])];
            tensor<fp16, [2, 640, 32, 32]> input_443_cast = reshape(shape = var_12761, x = hidden_states_273_cast)[name = tensor<string, []>("input_443_cast")];
            tensor<int32, [2]> var_12765 = const()[name = tensor<string, []>("op_12765"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12767 = const()[name = tensor<string, []>("op_12767"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211197952))), lut = tensor<fp16, [4]>([-0x1.a74p-4, -0x1.f9cp-6, 0x1.f14p-6, 0x1.a4cp-4]), 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, []>(211300416)))];
            tensor<fp16, [2, 640, 32, 32]> hidden_states_275_cast = conv(bias = up_blocks_2_attentions_2_proj_out_bias_to_fp16, dilations = var_12767, groups = var_10257, pad = hidden_states_275_pad_0, pad_type = hidden_states_275_pad_type_0, strides = var_12765, 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_12776 = const()[name = tensor<string, []>("op_12776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12778 = const()[name = tensor<string, []>("op_12778"), 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, [921600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211301760))), lut = tensor<fp16, [4]>([-0x1.9a8p-5, -0x1.cbp-7, 0x1.c68p-7, 0x1.99p-5]), 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, []>(212223424)))];
            tensor<fp16, [2, 640, 64, 64]> hidden_states_277_cast = conv(bias = up_blocks_2_upsamplers_0_conv_bias_to_fp16, dilations = var_12778, groups = var_10257, pad = hidden_states_277_pad_0, pad_type = hidden_states_277_pad_type_0, strides = var_12776, 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_12782 = const()[name = tensor<string, []>("op_12782"), val = tensor<int32, []>(3)];
            tensor<bool, []> var_12802 = const()[name = tensor<string, []>("op_12802"), val = tensor<bool, []>(true)];
            tensor<int32, []> var_12807 = const()[name = tensor<string, []>("op_12807"), 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_12807, 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, []>(212224768)))];
            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, []>(212226752)))];
            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_12834 = const()[name = tensor<string, []>("op_12834"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12836 = const()[name = tensor<string, []>("op_12836"), 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, [691200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212228736))), lut = tensor<fp16, [4]>([-0x1.f14p-4, -0x1.bep-6, 0x1.7dp-6, 0x1.ccp-4]), 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, []>(212920000)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_279_cast = conv(bias = up_blocks_3_resnets_0_conv1_bias_to_fp16, dilations = var_12836, groups = var_12807, pad = hidden_states_279_pad_0, pad_type = hidden_states_279_pad_type_0, strides = var_12834, 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_12842 = const()[name = tensor<string, []>("op_12842"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12844 = const()[name = tensor<string, []>("op_12844"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212920704))), lut = tensor<fp16, [4]>([-0x1.858p-5, -0x1.024p-8, 0x1p-8, 0x1.cf4p-5]), 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, []>(213023168)))];
            tensor<fp16, [2, 320, 1, 1]> temb_39_cast = conv(bias = up_blocks_3_resnets_0_time_emb_proj_bias_to_fp16, dilations = var_12844, groups = var_12807, pad = temb_39_pad_0, pad_type = temb_39_pad_type_0, strides = var_12842, 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, []>(213023872)))];
            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, []>(213024576)))];
            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_12854 = const()[name = tensor<string, []>("op_12854"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12856 = const()[name = tensor<string, []>("op_12856"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213025280))), lut = tensor<fp16, [4]>([-0x1.f78p-4, -0x1.cd8p-6, 0x1.dc4p-6, 0x1.004p-3]), 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, []>(213255744)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_281_cast = conv(bias = up_blocks_3_resnets_0_conv2_bias_to_fp16, dilations = var_12856, groups = var_12807, pad = hidden_states_281_pad_0, pad_type = hidden_states_281_pad_type_0, strides = var_12854, 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_12861 = const()[name = tensor<string, []>("op_12861"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12863 = const()[name = tensor<string, []>("op_12863"), 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, [76800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213256448))), lut = tensor<fp16, [4]>([-0x1.e9p-5, -0x1.36cp-7, 0x1.5dcp-7, 0x1.f7cp-5]), 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, []>(213333312)))];
            tensor<fp16, [2, 320, 64, 64]> x_23_cast = conv(bias = up_blocks_3_resnets_0_conv_shortcut_bias_to_fp16, dilations = var_12863, groups = var_12807, pad = x_23_pad_0, pad_type = x_23_pad_type_0, strides = var_12861, 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, []>(213334016)))];
            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, []>(213334720)))];
            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_12883 = const()[name = tensor<string, []>("op_12883"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12885 = const()[name = tensor<string, []>("op_12885"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213335424))), lut = tensor<fp16, [4]>([-0x1.a24p-4, -0x1.e44p-6, 0x1.ea4p-6, 0x1.a14p-4]), 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, []>(213361088)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_285_cast = conv(bias = up_blocks_3_attentions_0_proj_in_bias_to_fp16, dilations = var_12885, groups = var_12807, pad = hidden_states_285_pad_0, pad_type = hidden_states_285_pad_type_0, strides = var_12883, 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_12890 = const()[name = tensor<string, []>("op_12890"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_79_cast = reshape(shape = var_12890, x = hidden_states_285_cast)[name = tensor<string, []>("inputs_79_cast")];
            tensor<int32, [1]> var_12900 = const()[name = tensor<string, []>("op_12900"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_79_cast = reduce_mean(axes = var_12900, keep_dims = var_12802, 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_12904 = const()[name = tensor<string, []>("op_12904"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_12905_cast = reduce_mean(axes = var_12904, keep_dims = var_12802, x = zero_mean_sq_79_cast)[name = tensor<string, []>("op_12905_cast")];
            tensor<fp16, []> var_12906_to_fp16 = const()[name = tensor<string, []>("op_12906_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_12907_cast = add(x = var_12905_cast, y = var_12906_to_fp16)[name = tensor<string, []>("op_12907_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_12907_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_12911_to_fp16 = const()[name = tensor<string, []>("op_12911_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213361792)))];
            tensor<fp16, [2, 320, 1, 4096]> var_12912_cast = add(x = out_79_cast, y = var_12911_to_fp16)[name = tensor<string, []>("op_12912_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_12914_to_fp16 = const()[name = tensor<string, []>("op_12914_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213362496)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_287_cast = mul(x = var_12912_cast, y = var_12914_to_fp16)[name = tensor<string, []>("hidden_states_287_cast")];
            tensor<int32, [2]> var_12921 = const()[name = tensor<string, []>("op_12921"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12923 = const()[name = tensor<string, []>("op_12923"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213363200))), lut = tensor<fp16, [4]>([-0x1.288p-3, -0x1.6dcp-5, 0x1.4f4p-5, 0x1.22cp-3]), 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_12923, groups = var_12807, pad = q_53_pad_0, pad_type = q_53_pad_type_0, strides = var_12921, 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_12927 = const()[name = tensor<string, []>("op_12927"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12929 = const()[name = tensor<string, []>("op_12929"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213388864))), lut = tensor<fp16, [4]>([-0x1.19cp-3, -0x1.478p-5, 0x1.52p-5, 0x1.1ecp-3]), 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_12929, groups = var_12807, pad = k_105_pad_0, pad_type = k_105_pad_type_0, strides = var_12927, 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_12933 = const()[name = tensor<string, []>("op_12933"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_12935 = const()[name = tensor<string, []>("op_12935"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213414528))), lut = tensor<fp16, [4]>([-0x1.c6p-4, -0x1.024p-5, 0x1.01cp-5, 0x1.c18p-4]), 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_12935, groups = var_12807, pad = v_53_pad_0, pad_type = v_53_pad_type_0, strides = var_12933, 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_12939_begin_0 = const()[name = tensor<string, []>("op_12939_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12939_end_0 = const()[name = tensor<string, []>("op_12939_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_12939_end_mask_0 = const()[name = tensor<string, []>("op_12939_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_12939_cast = slice_by_index(begin = var_12939_begin_0, end = var_12939_end_0, end_mask = var_12939_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_12939_cast")];
            tensor<int32, [4]> var_12943_begin_0 = const()[name = tensor<string, []>("op_12943_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_12943_end_0 = const()[name = tensor<string, []>("op_12943_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_12943_end_mask_0 = const()[name = tensor<string, []>("op_12943_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_12943_cast = slice_by_index(begin = var_12943_begin_0, end = var_12943_end_0, end_mask = var_12943_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_12943_cast")];
            tensor<int32, [4]> var_12947_begin_0 = const()[name = tensor<string, []>("op_12947_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_12947_end_0 = const()[name = tensor<string, []>("op_12947_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_12947_end_mask_0 = const()[name = tensor<string, []>("op_12947_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_12947_cast = slice_by_index(begin = var_12947_begin_0, end = var_12947_end_0, end_mask = var_12947_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_12947_cast")];
            tensor<int32, [4]> var_12951_begin_0 = const()[name = tensor<string, []>("op_12951_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_12951_end_0 = const()[name = tensor<string, []>("op_12951_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_12951_end_mask_0 = const()[name = tensor<string, []>("op_12951_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_12951_cast = slice_by_index(begin = var_12951_begin_0, end = var_12951_end_0, end_mask = var_12951_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_12951_cast")];
            tensor<int32, [4]> var_12955_begin_0 = const()[name = tensor<string, []>("op_12955_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_12955_end_0 = const()[name = tensor<string, []>("op_12955_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_12955_end_mask_0 = const()[name = tensor<string, []>("op_12955_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_12955_cast = slice_by_index(begin = var_12955_begin_0, end = var_12955_end_0, end_mask = var_12955_end_mask_0, x = q_53_cast)[name = tensor<string, []>("op_12955_cast")];
            tensor<int32, [4]> var_12958_begin_0 = const()[name = tensor<string, []>("op_12958_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12958_end_0 = const()[name = tensor<string, []>("op_12958_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12958_end_mask_0 = const()[name = tensor<string, []>("op_12958_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12958_cast = slice_by_index(begin = var_12958_begin_0, end = var_12958_end_0, end_mask = var_12958_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12958_cast")];
            tensor<int32, [4]> var_12959_begin_0 = const()[name = tensor<string, []>("op_12959_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12959_end_0 = const()[name = tensor<string, []>("op_12959_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12959_end_mask_0 = const()[name = tensor<string, []>("op_12959_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12959_cast = slice_by_index(begin = var_12959_begin_0, end = var_12959_end_0, end_mask = var_12959_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12959_cast")];
            tensor<int32, [4]> var_12960_begin_0 = const()[name = tensor<string, []>("op_12960_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12960_end_0 = const()[name = tensor<string, []>("op_12960_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_12960_end_mask_0 = const()[name = tensor<string, []>("op_12960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12960_cast = slice_by_index(begin = var_12960_begin_0, end = var_12960_end_0, end_mask = var_12960_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12960_cast")];
            tensor<int32, [4]> var_12961_begin_0 = const()[name = tensor<string, []>("op_12961_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12961_end_0 = const()[name = tensor<string, []>("op_12961_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_12961_end_mask_0 = const()[name = tensor<string, []>("op_12961_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12961_cast = slice_by_index(begin = var_12961_begin_0, end = var_12961_end_0, end_mask = var_12961_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12961_cast")];
            tensor<int32, [4]> var_12962_begin_0 = const()[name = tensor<string, []>("op_12962_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12962_end_0 = const()[name = tensor<string, []>("op_12962_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_12962_end_mask_0 = const()[name = tensor<string, []>("op_12962_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12962_cast = slice_by_index(begin = var_12962_begin_0, end = var_12962_end_0, end_mask = var_12962_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12962_cast")];
            tensor<int32, [4]> var_12963_begin_0 = const()[name = tensor<string, []>("op_12963_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12963_end_0 = const()[name = tensor<string, []>("op_12963_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_12963_end_mask_0 = const()[name = tensor<string, []>("op_12963_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12963_cast = slice_by_index(begin = var_12963_begin_0, end = var_12963_end_0, end_mask = var_12963_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12963_cast")];
            tensor<int32, [4]> var_12964_begin_0 = const()[name = tensor<string, []>("op_12964_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12964_end_0 = const()[name = tensor<string, []>("op_12964_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_12964_end_mask_0 = const()[name = tensor<string, []>("op_12964_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12964_cast = slice_by_index(begin = var_12964_begin_0, end = var_12964_end_0, end_mask = var_12964_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12964_cast")];
            tensor<int32, [4]> var_12965_begin_0 = const()[name = tensor<string, []>("op_12965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12965_end_0 = const()[name = tensor<string, []>("op_12965_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_12965_end_mask_0 = const()[name = tensor<string, []>("op_12965_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12965_cast = slice_by_index(begin = var_12965_begin_0, end = var_12965_end_0, end_mask = var_12965_end_mask_0, x = var_12939_cast)[name = tensor<string, []>("op_12965_cast")];
            tensor<int32, [4]> var_12966_begin_0 = const()[name = tensor<string, []>("op_12966_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12966_end_0 = const()[name = tensor<string, []>("op_12966_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12966_end_mask_0 = const()[name = tensor<string, []>("op_12966_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12966_cast = slice_by_index(begin = var_12966_begin_0, end = var_12966_end_0, end_mask = var_12966_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12966_cast")];
            tensor<int32, [4]> var_12967_begin_0 = const()[name = tensor<string, []>("op_12967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12967_end_0 = const()[name = tensor<string, []>("op_12967_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12967_end_mask_0 = const()[name = tensor<string, []>("op_12967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12967_cast = slice_by_index(begin = var_12967_begin_0, end = var_12967_end_0, end_mask = var_12967_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12967_cast")];
            tensor<int32, [4]> var_12968_begin_0 = const()[name = tensor<string, []>("op_12968_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12968_end_0 = const()[name = tensor<string, []>("op_12968_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_12968_end_mask_0 = const()[name = tensor<string, []>("op_12968_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12968_cast = slice_by_index(begin = var_12968_begin_0, end = var_12968_end_0, end_mask = var_12968_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12968_cast")];
            tensor<int32, [4]> var_12969_begin_0 = const()[name = tensor<string, []>("op_12969_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12969_end_0 = const()[name = tensor<string, []>("op_12969_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_12969_end_mask_0 = const()[name = tensor<string, []>("op_12969_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12969_cast = slice_by_index(begin = var_12969_begin_0, end = var_12969_end_0, end_mask = var_12969_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12969_cast")];
            tensor<int32, [4]> var_12970_begin_0 = const()[name = tensor<string, []>("op_12970_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12970_end_0 = const()[name = tensor<string, []>("op_12970_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_12970_end_mask_0 = const()[name = tensor<string, []>("op_12970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12970_cast = slice_by_index(begin = var_12970_begin_0, end = var_12970_end_0, end_mask = var_12970_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12970_cast")];
            tensor<int32, [4]> var_12971_begin_0 = const()[name = tensor<string, []>("op_12971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12971_end_0 = const()[name = tensor<string, []>("op_12971_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_12971_end_mask_0 = const()[name = tensor<string, []>("op_12971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12971_cast = slice_by_index(begin = var_12971_begin_0, end = var_12971_end_0, end_mask = var_12971_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12971_cast")];
            tensor<int32, [4]> var_12972_begin_0 = const()[name = tensor<string, []>("op_12972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12972_end_0 = const()[name = tensor<string, []>("op_12972_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_12972_end_mask_0 = const()[name = tensor<string, []>("op_12972_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12972_cast = slice_by_index(begin = var_12972_begin_0, end = var_12972_end_0, end_mask = var_12972_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12972_cast")];
            tensor<int32, [4]> var_12973_begin_0 = const()[name = tensor<string, []>("op_12973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12973_end_0 = const()[name = tensor<string, []>("op_12973_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_12973_end_mask_0 = const()[name = tensor<string, []>("op_12973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12973_cast = slice_by_index(begin = var_12973_begin_0, end = var_12973_end_0, end_mask = var_12973_end_mask_0, x = var_12943_cast)[name = tensor<string, []>("op_12973_cast")];
            tensor<int32, [4]> var_12974_begin_0 = const()[name = tensor<string, []>("op_12974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12974_end_0 = const()[name = tensor<string, []>("op_12974_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12974_end_mask_0 = const()[name = tensor<string, []>("op_12974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12974_cast = slice_by_index(begin = var_12974_begin_0, end = var_12974_end_0, end_mask = var_12974_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12974_cast")];
            tensor<int32, [4]> var_12975_begin_0 = const()[name = tensor<string, []>("op_12975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12975_end_0 = const()[name = tensor<string, []>("op_12975_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12975_end_mask_0 = const()[name = tensor<string, []>("op_12975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12975_cast = slice_by_index(begin = var_12975_begin_0, end = var_12975_end_0, end_mask = var_12975_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12975_cast")];
            tensor<int32, [4]> var_12976_begin_0 = const()[name = tensor<string, []>("op_12976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12976_end_0 = const()[name = tensor<string, []>("op_12976_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_12976_end_mask_0 = const()[name = tensor<string, []>("op_12976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12976_cast = slice_by_index(begin = var_12976_begin_0, end = var_12976_end_0, end_mask = var_12976_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12976_cast")];
            tensor<int32, [4]> var_12977_begin_0 = const()[name = tensor<string, []>("op_12977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12977_end_0 = const()[name = tensor<string, []>("op_12977_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_12977_end_mask_0 = const()[name = tensor<string, []>("op_12977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12977_cast = slice_by_index(begin = var_12977_begin_0, end = var_12977_end_0, end_mask = var_12977_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12977_cast")];
            tensor<int32, [4]> var_12978_begin_0 = const()[name = tensor<string, []>("op_12978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12978_end_0 = const()[name = tensor<string, []>("op_12978_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_12978_end_mask_0 = const()[name = tensor<string, []>("op_12978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12978_cast = slice_by_index(begin = var_12978_begin_0, end = var_12978_end_0, end_mask = var_12978_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12978_cast")];
            tensor<int32, [4]> var_12979_begin_0 = const()[name = tensor<string, []>("op_12979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12979_end_0 = const()[name = tensor<string, []>("op_12979_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_12979_end_mask_0 = const()[name = tensor<string, []>("op_12979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12979_cast = slice_by_index(begin = var_12979_begin_0, end = var_12979_end_0, end_mask = var_12979_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12979_cast")];
            tensor<int32, [4]> var_12980_begin_0 = const()[name = tensor<string, []>("op_12980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12980_end_0 = const()[name = tensor<string, []>("op_12980_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_12980_end_mask_0 = const()[name = tensor<string, []>("op_12980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12980_cast = slice_by_index(begin = var_12980_begin_0, end = var_12980_end_0, end_mask = var_12980_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12980_cast")];
            tensor<int32, [4]> var_12981_begin_0 = const()[name = tensor<string, []>("op_12981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12981_end_0 = const()[name = tensor<string, []>("op_12981_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_12981_end_mask_0 = const()[name = tensor<string, []>("op_12981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12981_cast = slice_by_index(begin = var_12981_begin_0, end = var_12981_end_0, end_mask = var_12981_end_mask_0, x = var_12947_cast)[name = tensor<string, []>("op_12981_cast")];
            tensor<int32, [4]> var_12982_begin_0 = const()[name = tensor<string, []>("op_12982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12982_end_0 = const()[name = tensor<string, []>("op_12982_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12982_end_mask_0 = const()[name = tensor<string, []>("op_12982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12982_cast = slice_by_index(begin = var_12982_begin_0, end = var_12982_end_0, end_mask = var_12982_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12982_cast")];
            tensor<int32, [4]> var_12983_begin_0 = const()[name = tensor<string, []>("op_12983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12983_end_0 = const()[name = tensor<string, []>("op_12983_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12983_end_mask_0 = const()[name = tensor<string, []>("op_12983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12983_cast = slice_by_index(begin = var_12983_begin_0, end = var_12983_end_0, end_mask = var_12983_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12983_cast")];
            tensor<int32, [4]> var_12984_begin_0 = const()[name = tensor<string, []>("op_12984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12984_end_0 = const()[name = tensor<string, []>("op_12984_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_12984_end_mask_0 = const()[name = tensor<string, []>("op_12984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12984_cast = slice_by_index(begin = var_12984_begin_0, end = var_12984_end_0, end_mask = var_12984_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12984_cast")];
            tensor<int32, [4]> var_12985_begin_0 = const()[name = tensor<string, []>("op_12985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12985_end_0 = const()[name = tensor<string, []>("op_12985_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_12985_end_mask_0 = const()[name = tensor<string, []>("op_12985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12985_cast = slice_by_index(begin = var_12985_begin_0, end = var_12985_end_0, end_mask = var_12985_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12985_cast")];
            tensor<int32, [4]> var_12986_begin_0 = const()[name = tensor<string, []>("op_12986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12986_end_0 = const()[name = tensor<string, []>("op_12986_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_12986_end_mask_0 = const()[name = tensor<string, []>("op_12986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12986_cast = slice_by_index(begin = var_12986_begin_0, end = var_12986_end_0, end_mask = var_12986_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12986_cast")];
            tensor<int32, [4]> var_12987_begin_0 = const()[name = tensor<string, []>("op_12987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12987_end_0 = const()[name = tensor<string, []>("op_12987_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_12987_end_mask_0 = const()[name = tensor<string, []>("op_12987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12987_cast = slice_by_index(begin = var_12987_begin_0, end = var_12987_end_0, end_mask = var_12987_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12987_cast")];
            tensor<int32, [4]> var_12988_begin_0 = const()[name = tensor<string, []>("op_12988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12988_end_0 = const()[name = tensor<string, []>("op_12988_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_12988_end_mask_0 = const()[name = tensor<string, []>("op_12988_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12988_cast = slice_by_index(begin = var_12988_begin_0, end = var_12988_end_0, end_mask = var_12988_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12988_cast")];
            tensor<int32, [4]> var_12989_begin_0 = const()[name = tensor<string, []>("op_12989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12989_end_0 = const()[name = tensor<string, []>("op_12989_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_12989_end_mask_0 = const()[name = tensor<string, []>("op_12989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12989_cast = slice_by_index(begin = var_12989_begin_0, end = var_12989_end_0, end_mask = var_12989_end_mask_0, x = var_12951_cast)[name = tensor<string, []>("op_12989_cast")];
            tensor<int32, [4]> var_12990_begin_0 = const()[name = tensor<string, []>("op_12990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_12990_end_0 = const()[name = tensor<string, []>("op_12990_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_12990_end_mask_0 = const()[name = tensor<string, []>("op_12990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12990_cast = slice_by_index(begin = var_12990_begin_0, end = var_12990_end_0, end_mask = var_12990_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12990_cast")];
            tensor<int32, [4]> var_12991_begin_0 = const()[name = tensor<string, []>("op_12991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_12991_end_0 = const()[name = tensor<string, []>("op_12991_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_12991_end_mask_0 = const()[name = tensor<string, []>("op_12991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12991_cast = slice_by_index(begin = var_12991_begin_0, end = var_12991_end_0, end_mask = var_12991_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12991_cast")];
            tensor<int32, [4]> var_12992_begin_0 = const()[name = tensor<string, []>("op_12992_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_12992_end_0 = const()[name = tensor<string, []>("op_12992_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_12992_end_mask_0 = const()[name = tensor<string, []>("op_12992_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12992_cast = slice_by_index(begin = var_12992_begin_0, end = var_12992_end_0, end_mask = var_12992_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12992_cast")];
            tensor<int32, [4]> var_12993_begin_0 = const()[name = tensor<string, []>("op_12993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_12993_end_0 = const()[name = tensor<string, []>("op_12993_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_12993_end_mask_0 = const()[name = tensor<string, []>("op_12993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12993_cast = slice_by_index(begin = var_12993_begin_0, end = var_12993_end_0, end_mask = var_12993_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12993_cast")];
            tensor<int32, [4]> var_12994_begin_0 = const()[name = tensor<string, []>("op_12994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_12994_end_0 = const()[name = tensor<string, []>("op_12994_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_12994_end_mask_0 = const()[name = tensor<string, []>("op_12994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12994_cast = slice_by_index(begin = var_12994_begin_0, end = var_12994_end_0, end_mask = var_12994_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12994_cast")];
            tensor<int32, [4]> var_12995_begin_0 = const()[name = tensor<string, []>("op_12995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_12995_end_0 = const()[name = tensor<string, []>("op_12995_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_12995_end_mask_0 = const()[name = tensor<string, []>("op_12995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12995_cast = slice_by_index(begin = var_12995_begin_0, end = var_12995_end_0, end_mask = var_12995_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12995_cast")];
            tensor<int32, [4]> var_12996_begin_0 = const()[name = tensor<string, []>("op_12996_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_12996_end_0 = const()[name = tensor<string, []>("op_12996_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_12996_end_mask_0 = const()[name = tensor<string, []>("op_12996_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12996_cast = slice_by_index(begin = var_12996_begin_0, end = var_12996_end_0, end_mask = var_12996_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12996_cast")];
            tensor<int32, [4]> var_12997_begin_0 = const()[name = tensor<string, []>("op_12997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_12997_end_0 = const()[name = tensor<string, []>("op_12997_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_12997_end_mask_0 = const()[name = tensor<string, []>("op_12997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_12997_cast = slice_by_index(begin = var_12997_begin_0, end = var_12997_end_0, end_mask = var_12997_end_mask_0, x = var_12955_cast)[name = tensor<string, []>("op_12997_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_13002_begin_0 = const()[name = tensor<string, []>("op_13002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13002_end_0 = const()[name = tensor<string, []>("op_13002_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_13002_end_mask_0 = const()[name = tensor<string, []>("op_13002_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, 64]> var_13002_cast = slice_by_index(begin = var_13002_begin_0, end = var_13002_end_0, end_mask = var_13002_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_13002_cast")];
            tensor<int32, [4]> var_13006_begin_0 = const()[name = tensor<string, []>("op_13006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_13006_end_0 = const()[name = tensor<string, []>("op_13006_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_13006_end_mask_0 = const()[name = tensor<string, []>("op_13006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_13006_cast = slice_by_index(begin = var_13006_begin_0, end = var_13006_end_0, end_mask = var_13006_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_13006_cast")];
            tensor<int32, [4]> var_13010_begin_0 = const()[name = tensor<string, []>("op_13010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_13010_end_0 = const()[name = tensor<string, []>("op_13010_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_13010_end_mask_0 = const()[name = tensor<string, []>("op_13010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_13010_cast = slice_by_index(begin = var_13010_begin_0, end = var_13010_end_0, end_mask = var_13010_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_13010_cast")];
            tensor<int32, [4]> var_13014_begin_0 = const()[name = tensor<string, []>("op_13014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_13014_end_0 = const()[name = tensor<string, []>("op_13014_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_13014_end_mask_0 = const()[name = tensor<string, []>("op_13014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_13014_cast = slice_by_index(begin = var_13014_begin_0, end = var_13014_end_0, end_mask = var_13014_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_13014_cast")];
            tensor<int32, [4]> var_13018_begin_0 = const()[name = tensor<string, []>("op_13018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_13018_end_0 = const()[name = tensor<string, []>("op_13018_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_13018_end_mask_0 = const()[name = tensor<string, []>("op_13018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_13018_cast = slice_by_index(begin = var_13018_begin_0, end = var_13018_end_0, end_mask = var_13018_end_mask_0, x = transpose_5)[name = tensor<string, []>("op_13018_cast")];
            tensor<int32, [4]> var_13020_begin_0 = const()[name = tensor<string, []>("op_13020_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13020_end_0 = const()[name = tensor<string, []>("op_13020_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13020_end_mask_0 = const()[name = tensor<string, []>("op_13020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13020_cast = slice_by_index(begin = var_13020_begin_0, end = var_13020_end_0, end_mask = var_13020_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_13020_cast")];
            tensor<int32, [4]> var_13024_begin_0 = const()[name = tensor<string, []>("op_13024_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13024_end_0 = const()[name = tensor<string, []>("op_13024_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_13024_end_mask_0 = const()[name = tensor<string, []>("op_13024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13024_cast = slice_by_index(begin = var_13024_begin_0, end = var_13024_end_0, end_mask = var_13024_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_13024_cast")];
            tensor<int32, [4]> var_13028_begin_0 = const()[name = tensor<string, []>("op_13028_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13028_end_0 = const()[name = tensor<string, []>("op_13028_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_13028_end_mask_0 = const()[name = tensor<string, []>("op_13028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13028_cast = slice_by_index(begin = var_13028_begin_0, end = var_13028_end_0, end_mask = var_13028_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_13028_cast")];
            tensor<int32, [4]> var_13032_begin_0 = const()[name = tensor<string, []>("op_13032_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13032_end_0 = const()[name = tensor<string, []>("op_13032_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_13032_end_mask_0 = const()[name = tensor<string, []>("op_13032_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13032_cast = slice_by_index(begin = var_13032_begin_0, end = var_13032_end_0, end_mask = var_13032_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_13032_cast")];
            tensor<int32, [4]> var_13036_begin_0 = const()[name = tensor<string, []>("op_13036_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13036_end_0 = const()[name = tensor<string, []>("op_13036_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_13036_end_mask_0 = const()[name = tensor<string, []>("op_13036_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13036_cast = slice_by_index(begin = var_13036_begin_0, end = var_13036_end_0, end_mask = var_13036_end_mask_0, x = v_53_cast)[name = tensor<string, []>("op_13036_cast")];
            tensor<string, []> var_13040_equation_0 = const()[name = tensor<string, []>("op_13040_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13040_cast = einsum(equation = var_13040_equation_0, values = (var_13002_cast, var_12958_cast))[name = tensor<string, []>("op_13040_cast")];
            tensor<fp16, []> var_13041_to_fp16 = const()[name = tensor<string, []>("op_13041_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_721_cast = mul(x = var_13040_cast, y = var_13041_to_fp16)[name = tensor<string, []>("aw_chunk_721_cast")];
            tensor<string, []> var_13044_equation_0 = const()[name = tensor<string, []>("op_13044_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13044_cast = einsum(equation = var_13044_equation_0, values = (var_13002_cast, var_12959_cast))[name = tensor<string, []>("op_13044_cast")];
            tensor<fp16, []> var_13045_to_fp16 = const()[name = tensor<string, []>("op_13045_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_723_cast = mul(x = var_13044_cast, y = var_13045_to_fp16)[name = tensor<string, []>("aw_chunk_723_cast")];
            tensor<string, []> var_13048_equation_0 = const()[name = tensor<string, []>("op_13048_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13048_cast = einsum(equation = var_13048_equation_0, values = (var_13002_cast, var_12960_cast))[name = tensor<string, []>("op_13048_cast")];
            tensor<fp16, []> var_13049_to_fp16 = const()[name = tensor<string, []>("op_13049_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_725_cast = mul(x = var_13048_cast, y = var_13049_to_fp16)[name = tensor<string, []>("aw_chunk_725_cast")];
            tensor<string, []> var_13052_equation_0 = const()[name = tensor<string, []>("op_13052_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13052_cast = einsum(equation = var_13052_equation_0, values = (var_13002_cast, var_12961_cast))[name = tensor<string, []>("op_13052_cast")];
            tensor<fp16, []> var_13053_to_fp16 = const()[name = tensor<string, []>("op_13053_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_727_cast = mul(x = var_13052_cast, y = var_13053_to_fp16)[name = tensor<string, []>("aw_chunk_727_cast")];
            tensor<string, []> var_13056_equation_0 = const()[name = tensor<string, []>("op_13056_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13056_cast = einsum(equation = var_13056_equation_0, values = (var_13002_cast, var_12962_cast))[name = tensor<string, []>("op_13056_cast")];
            tensor<fp16, []> var_13057_to_fp16 = const()[name = tensor<string, []>("op_13057_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_729_cast = mul(x = var_13056_cast, y = var_13057_to_fp16)[name = tensor<string, []>("aw_chunk_729_cast")];
            tensor<string, []> var_13060_equation_0 = const()[name = tensor<string, []>("op_13060_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13060_cast = einsum(equation = var_13060_equation_0, values = (var_13002_cast, var_12963_cast))[name = tensor<string, []>("op_13060_cast")];
            tensor<fp16, []> var_13061_to_fp16 = const()[name = tensor<string, []>("op_13061_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_731_cast = mul(x = var_13060_cast, y = var_13061_to_fp16)[name = tensor<string, []>("aw_chunk_731_cast")];
            tensor<string, []> var_13064_equation_0 = const()[name = tensor<string, []>("op_13064_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13064_cast = einsum(equation = var_13064_equation_0, values = (var_13002_cast, var_12964_cast))[name = tensor<string, []>("op_13064_cast")];
            tensor<fp16, []> var_13065_to_fp16 = const()[name = tensor<string, []>("op_13065_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_733_cast = mul(x = var_13064_cast, y = var_13065_to_fp16)[name = tensor<string, []>("aw_chunk_733_cast")];
            tensor<string, []> var_13068_equation_0 = const()[name = tensor<string, []>("op_13068_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13068_cast = einsum(equation = var_13068_equation_0, values = (var_13002_cast, var_12965_cast))[name = tensor<string, []>("op_13068_cast")];
            tensor<fp16, []> var_13069_to_fp16 = const()[name = tensor<string, []>("op_13069_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_735_cast = mul(x = var_13068_cast, y = var_13069_to_fp16)[name = tensor<string, []>("aw_chunk_735_cast")];
            tensor<string, []> var_13072_equation_0 = const()[name = tensor<string, []>("op_13072_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13072_cast = einsum(equation = var_13072_equation_0, values = (var_13006_cast, var_12966_cast))[name = tensor<string, []>("op_13072_cast")];
            tensor<fp16, []> var_13073_to_fp16 = const()[name = tensor<string, []>("op_13073_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_737_cast = mul(x = var_13072_cast, y = var_13073_to_fp16)[name = tensor<string, []>("aw_chunk_737_cast")];
            tensor<string, []> var_13076_equation_0 = const()[name = tensor<string, []>("op_13076_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13076_cast = einsum(equation = var_13076_equation_0, values = (var_13006_cast, var_12967_cast))[name = tensor<string, []>("op_13076_cast")];
            tensor<fp16, []> var_13077_to_fp16 = const()[name = tensor<string, []>("op_13077_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_739_cast = mul(x = var_13076_cast, y = var_13077_to_fp16)[name = tensor<string, []>("aw_chunk_739_cast")];
            tensor<string, []> var_13080_equation_0 = const()[name = tensor<string, []>("op_13080_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13080_cast = einsum(equation = var_13080_equation_0, values = (var_13006_cast, var_12968_cast))[name = tensor<string, []>("op_13080_cast")];
            tensor<fp16, []> var_13081_to_fp16 = const()[name = tensor<string, []>("op_13081_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_741_cast = mul(x = var_13080_cast, y = var_13081_to_fp16)[name = tensor<string, []>("aw_chunk_741_cast")];
            tensor<string, []> var_13084_equation_0 = const()[name = tensor<string, []>("op_13084_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13084_cast = einsum(equation = var_13084_equation_0, values = (var_13006_cast, var_12969_cast))[name = tensor<string, []>("op_13084_cast")];
            tensor<fp16, []> var_13085_to_fp16 = const()[name = tensor<string, []>("op_13085_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_743_cast = mul(x = var_13084_cast, y = var_13085_to_fp16)[name = tensor<string, []>("aw_chunk_743_cast")];
            tensor<string, []> var_13088_equation_0 = const()[name = tensor<string, []>("op_13088_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13088_cast = einsum(equation = var_13088_equation_0, values = (var_13006_cast, var_12970_cast))[name = tensor<string, []>("op_13088_cast")];
            tensor<fp16, []> var_13089_to_fp16 = const()[name = tensor<string, []>("op_13089_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_745_cast = mul(x = var_13088_cast, y = var_13089_to_fp16)[name = tensor<string, []>("aw_chunk_745_cast")];
            tensor<string, []> var_13092_equation_0 = const()[name = tensor<string, []>("op_13092_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13092_cast = einsum(equation = var_13092_equation_0, values = (var_13006_cast, var_12971_cast))[name = tensor<string, []>("op_13092_cast")];
            tensor<fp16, []> var_13093_to_fp16 = const()[name = tensor<string, []>("op_13093_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_747_cast = mul(x = var_13092_cast, y = var_13093_to_fp16)[name = tensor<string, []>("aw_chunk_747_cast")];
            tensor<string, []> var_13096_equation_0 = const()[name = tensor<string, []>("op_13096_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13096_cast = einsum(equation = var_13096_equation_0, values = (var_13006_cast, var_12972_cast))[name = tensor<string, []>("op_13096_cast")];
            tensor<fp16, []> var_13097_to_fp16 = const()[name = tensor<string, []>("op_13097_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_749_cast = mul(x = var_13096_cast, y = var_13097_to_fp16)[name = tensor<string, []>("aw_chunk_749_cast")];
            tensor<string, []> var_13100_equation_0 = const()[name = tensor<string, []>("op_13100_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13100_cast = einsum(equation = var_13100_equation_0, values = (var_13006_cast, var_12973_cast))[name = tensor<string, []>("op_13100_cast")];
            tensor<fp16, []> var_13101_to_fp16 = const()[name = tensor<string, []>("op_13101_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_751_cast = mul(x = var_13100_cast, y = var_13101_to_fp16)[name = tensor<string, []>("aw_chunk_751_cast")];
            tensor<string, []> var_13104_equation_0 = const()[name = tensor<string, []>("op_13104_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13104_cast = einsum(equation = var_13104_equation_0, values = (var_13010_cast, var_12974_cast))[name = tensor<string, []>("op_13104_cast")];
            tensor<fp16, []> var_13105_to_fp16 = const()[name = tensor<string, []>("op_13105_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_753_cast = mul(x = var_13104_cast, y = var_13105_to_fp16)[name = tensor<string, []>("aw_chunk_753_cast")];
            tensor<string, []> var_13108_equation_0 = const()[name = tensor<string, []>("op_13108_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13108_cast = einsum(equation = var_13108_equation_0, values = (var_13010_cast, var_12975_cast))[name = tensor<string, []>("op_13108_cast")];
            tensor<fp16, []> var_13109_to_fp16 = const()[name = tensor<string, []>("op_13109_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_755_cast = mul(x = var_13108_cast, y = var_13109_to_fp16)[name = tensor<string, []>("aw_chunk_755_cast")];
            tensor<string, []> var_13112_equation_0 = const()[name = tensor<string, []>("op_13112_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13112_cast = einsum(equation = var_13112_equation_0, values = (var_13010_cast, var_12976_cast))[name = tensor<string, []>("op_13112_cast")];
            tensor<fp16, []> var_13113_to_fp16 = const()[name = tensor<string, []>("op_13113_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_757_cast = mul(x = var_13112_cast, y = var_13113_to_fp16)[name = tensor<string, []>("aw_chunk_757_cast")];
            tensor<string, []> var_13116_equation_0 = const()[name = tensor<string, []>("op_13116_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13116_cast = einsum(equation = var_13116_equation_0, values = (var_13010_cast, var_12977_cast))[name = tensor<string, []>("op_13116_cast")];
            tensor<fp16, []> var_13117_to_fp16 = const()[name = tensor<string, []>("op_13117_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_759_cast = mul(x = var_13116_cast, y = var_13117_to_fp16)[name = tensor<string, []>("aw_chunk_759_cast")];
            tensor<string, []> var_13120_equation_0 = const()[name = tensor<string, []>("op_13120_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13120_cast = einsum(equation = var_13120_equation_0, values = (var_13010_cast, var_12978_cast))[name = tensor<string, []>("op_13120_cast")];
            tensor<fp16, []> var_13121_to_fp16 = const()[name = tensor<string, []>("op_13121_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_761_cast = mul(x = var_13120_cast, y = var_13121_to_fp16)[name = tensor<string, []>("aw_chunk_761_cast")];
            tensor<string, []> var_13124_equation_0 = const()[name = tensor<string, []>("op_13124_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13124_cast = einsum(equation = var_13124_equation_0, values = (var_13010_cast, var_12979_cast))[name = tensor<string, []>("op_13124_cast")];
            tensor<fp16, []> var_13125_to_fp16 = const()[name = tensor<string, []>("op_13125_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_763_cast = mul(x = var_13124_cast, y = var_13125_to_fp16)[name = tensor<string, []>("aw_chunk_763_cast")];
            tensor<string, []> var_13128_equation_0 = const()[name = tensor<string, []>("op_13128_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13128_cast = einsum(equation = var_13128_equation_0, values = (var_13010_cast, var_12980_cast))[name = tensor<string, []>("op_13128_cast")];
            tensor<fp16, []> var_13129_to_fp16 = const()[name = tensor<string, []>("op_13129_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_765_cast = mul(x = var_13128_cast, y = var_13129_to_fp16)[name = tensor<string, []>("aw_chunk_765_cast")];
            tensor<string, []> var_13132_equation_0 = const()[name = tensor<string, []>("op_13132_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13132_cast = einsum(equation = var_13132_equation_0, values = (var_13010_cast, var_12981_cast))[name = tensor<string, []>("op_13132_cast")];
            tensor<fp16, []> var_13133_to_fp16 = const()[name = tensor<string, []>("op_13133_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_767_cast = mul(x = var_13132_cast, y = var_13133_to_fp16)[name = tensor<string, []>("aw_chunk_767_cast")];
            tensor<string, []> var_13136_equation_0 = const()[name = tensor<string, []>("op_13136_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13136_cast = einsum(equation = var_13136_equation_0, values = (var_13014_cast, var_12982_cast))[name = tensor<string, []>("op_13136_cast")];
            tensor<fp16, []> var_13137_to_fp16 = const()[name = tensor<string, []>("op_13137_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_769_cast = mul(x = var_13136_cast, y = var_13137_to_fp16)[name = tensor<string, []>("aw_chunk_769_cast")];
            tensor<string, []> var_13140_equation_0 = const()[name = tensor<string, []>("op_13140_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13140_cast = einsum(equation = var_13140_equation_0, values = (var_13014_cast, var_12983_cast))[name = tensor<string, []>("op_13140_cast")];
            tensor<fp16, []> var_13141_to_fp16 = const()[name = tensor<string, []>("op_13141_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_771_cast = mul(x = var_13140_cast, y = var_13141_to_fp16)[name = tensor<string, []>("aw_chunk_771_cast")];
            tensor<string, []> var_13144_equation_0 = const()[name = tensor<string, []>("op_13144_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13144_cast = einsum(equation = var_13144_equation_0, values = (var_13014_cast, var_12984_cast))[name = tensor<string, []>("op_13144_cast")];
            tensor<fp16, []> var_13145_to_fp16 = const()[name = tensor<string, []>("op_13145_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_773_cast = mul(x = var_13144_cast, y = var_13145_to_fp16)[name = tensor<string, []>("aw_chunk_773_cast")];
            tensor<string, []> var_13148_equation_0 = const()[name = tensor<string, []>("op_13148_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13148_cast = einsum(equation = var_13148_equation_0, values = (var_13014_cast, var_12985_cast))[name = tensor<string, []>("op_13148_cast")];
            tensor<fp16, []> var_13149_to_fp16 = const()[name = tensor<string, []>("op_13149_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_775_cast = mul(x = var_13148_cast, y = var_13149_to_fp16)[name = tensor<string, []>("aw_chunk_775_cast")];
            tensor<string, []> var_13152_equation_0 = const()[name = tensor<string, []>("op_13152_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13152_cast = einsum(equation = var_13152_equation_0, values = (var_13014_cast, var_12986_cast))[name = tensor<string, []>("op_13152_cast")];
            tensor<fp16, []> var_13153_to_fp16 = const()[name = tensor<string, []>("op_13153_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_777_cast = mul(x = var_13152_cast, y = var_13153_to_fp16)[name = tensor<string, []>("aw_chunk_777_cast")];
            tensor<string, []> var_13156_equation_0 = const()[name = tensor<string, []>("op_13156_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13156_cast = einsum(equation = var_13156_equation_0, values = (var_13014_cast, var_12987_cast))[name = tensor<string, []>("op_13156_cast")];
            tensor<fp16, []> var_13157_to_fp16 = const()[name = tensor<string, []>("op_13157_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_779_cast = mul(x = var_13156_cast, y = var_13157_to_fp16)[name = tensor<string, []>("aw_chunk_779_cast")];
            tensor<string, []> var_13160_equation_0 = const()[name = tensor<string, []>("op_13160_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13160_cast = einsum(equation = var_13160_equation_0, values = (var_13014_cast, var_12988_cast))[name = tensor<string, []>("op_13160_cast")];
            tensor<fp16, []> var_13161_to_fp16 = const()[name = tensor<string, []>("op_13161_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_781_cast = mul(x = var_13160_cast, y = var_13161_to_fp16)[name = tensor<string, []>("aw_chunk_781_cast")];
            tensor<string, []> var_13164_equation_0 = const()[name = tensor<string, []>("op_13164_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13164_cast = einsum(equation = var_13164_equation_0, values = (var_13014_cast, var_12989_cast))[name = tensor<string, []>("op_13164_cast")];
            tensor<fp16, []> var_13165_to_fp16 = const()[name = tensor<string, []>("op_13165_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_783_cast = mul(x = var_13164_cast, y = var_13165_to_fp16)[name = tensor<string, []>("aw_chunk_783_cast")];
            tensor<string, []> var_13168_equation_0 = const()[name = tensor<string, []>("op_13168_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13168_cast = einsum(equation = var_13168_equation_0, values = (var_13018_cast, var_12990_cast))[name = tensor<string, []>("op_13168_cast")];
            tensor<fp16, []> var_13169_to_fp16 = const()[name = tensor<string, []>("op_13169_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_785_cast = mul(x = var_13168_cast, y = var_13169_to_fp16)[name = tensor<string, []>("aw_chunk_785_cast")];
            tensor<string, []> var_13172_equation_0 = const()[name = tensor<string, []>("op_13172_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13172_cast = einsum(equation = var_13172_equation_0, values = (var_13018_cast, var_12991_cast))[name = tensor<string, []>("op_13172_cast")];
            tensor<fp16, []> var_13173_to_fp16 = const()[name = tensor<string, []>("op_13173_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_787_cast = mul(x = var_13172_cast, y = var_13173_to_fp16)[name = tensor<string, []>("aw_chunk_787_cast")];
            tensor<string, []> var_13176_equation_0 = const()[name = tensor<string, []>("op_13176_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13176_cast = einsum(equation = var_13176_equation_0, values = (var_13018_cast, var_12992_cast))[name = tensor<string, []>("op_13176_cast")];
            tensor<fp16, []> var_13177_to_fp16 = const()[name = tensor<string, []>("op_13177_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_789_cast = mul(x = var_13176_cast, y = var_13177_to_fp16)[name = tensor<string, []>("aw_chunk_789_cast")];
            tensor<string, []> var_13180_equation_0 = const()[name = tensor<string, []>("op_13180_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13180_cast = einsum(equation = var_13180_equation_0, values = (var_13018_cast, var_12993_cast))[name = tensor<string, []>("op_13180_cast")];
            tensor<fp16, []> var_13181_to_fp16 = const()[name = tensor<string, []>("op_13181_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_791_cast = mul(x = var_13180_cast, y = var_13181_to_fp16)[name = tensor<string, []>("aw_chunk_791_cast")];
            tensor<string, []> var_13184_equation_0 = const()[name = tensor<string, []>("op_13184_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13184_cast = einsum(equation = var_13184_equation_0, values = (var_13018_cast, var_12994_cast))[name = tensor<string, []>("op_13184_cast")];
            tensor<fp16, []> var_13185_to_fp16 = const()[name = tensor<string, []>("op_13185_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_793_cast = mul(x = var_13184_cast, y = var_13185_to_fp16)[name = tensor<string, []>("aw_chunk_793_cast")];
            tensor<string, []> var_13188_equation_0 = const()[name = tensor<string, []>("op_13188_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13188_cast = einsum(equation = var_13188_equation_0, values = (var_13018_cast, var_12995_cast))[name = tensor<string, []>("op_13188_cast")];
            tensor<fp16, []> var_13189_to_fp16 = const()[name = tensor<string, []>("op_13189_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_795_cast = mul(x = var_13188_cast, y = var_13189_to_fp16)[name = tensor<string, []>("aw_chunk_795_cast")];
            tensor<string, []> var_13192_equation_0 = const()[name = tensor<string, []>("op_13192_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13192_cast = einsum(equation = var_13192_equation_0, values = (var_13018_cast, var_12996_cast))[name = tensor<string, []>("op_13192_cast")];
            tensor<fp16, []> var_13193_to_fp16 = const()[name = tensor<string, []>("op_13193_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_797_cast = mul(x = var_13192_cast, y = var_13193_to_fp16)[name = tensor<string, []>("aw_chunk_797_cast")];
            tensor<string, []> var_13196_equation_0 = const()[name = tensor<string, []>("op_13196_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_13196_cast = einsum(equation = var_13196_equation_0, values = (var_13018_cast, var_12997_cast))[name = tensor<string, []>("op_13196_cast")];
            tensor<fp16, []> var_13197_to_fp16 = const()[name = tensor<string, []>("op_13197_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_799_cast = mul(x = var_13196_cast, y = var_13197_to_fp16)[name = tensor<string, []>("aw_chunk_799_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13199_cast = softmax(axis = var_12807, x = aw_chunk_721_cast)[name = tensor<string, []>("op_13199_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13200_cast = softmax(axis = var_12807, x = aw_chunk_723_cast)[name = tensor<string, []>("op_13200_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13201_cast = softmax(axis = var_12807, x = aw_chunk_725_cast)[name = tensor<string, []>("op_13201_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13202_cast = softmax(axis = var_12807, x = aw_chunk_727_cast)[name = tensor<string, []>("op_13202_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13203_cast = softmax(axis = var_12807, x = aw_chunk_729_cast)[name = tensor<string, []>("op_13203_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13204_cast = softmax(axis = var_12807, x = aw_chunk_731_cast)[name = tensor<string, []>("op_13204_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13205_cast = softmax(axis = var_12807, x = aw_chunk_733_cast)[name = tensor<string, []>("op_13205_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13206_cast = softmax(axis = var_12807, x = aw_chunk_735_cast)[name = tensor<string, []>("op_13206_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13207_cast = softmax(axis = var_12807, x = aw_chunk_737_cast)[name = tensor<string, []>("op_13207_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13208_cast = softmax(axis = var_12807, x = aw_chunk_739_cast)[name = tensor<string, []>("op_13208_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13209_cast = softmax(axis = var_12807, x = aw_chunk_741_cast)[name = tensor<string, []>("op_13209_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13210_cast = softmax(axis = var_12807, x = aw_chunk_743_cast)[name = tensor<string, []>("op_13210_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13211_cast = softmax(axis = var_12807, x = aw_chunk_745_cast)[name = tensor<string, []>("op_13211_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13212_cast = softmax(axis = var_12807, x = aw_chunk_747_cast)[name = tensor<string, []>("op_13212_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13213_cast = softmax(axis = var_12807, x = aw_chunk_749_cast)[name = tensor<string, []>("op_13213_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13214_cast = softmax(axis = var_12807, x = aw_chunk_751_cast)[name = tensor<string, []>("op_13214_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13215_cast = softmax(axis = var_12807, x = aw_chunk_753_cast)[name = tensor<string, []>("op_13215_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13216_cast = softmax(axis = var_12807, x = aw_chunk_755_cast)[name = tensor<string, []>("op_13216_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13217_cast = softmax(axis = var_12807, x = aw_chunk_757_cast)[name = tensor<string, []>("op_13217_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13218_cast = softmax(axis = var_12807, x = aw_chunk_759_cast)[name = tensor<string, []>("op_13218_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13219_cast = softmax(axis = var_12807, x = aw_chunk_761_cast)[name = tensor<string, []>("op_13219_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13220_cast = softmax(axis = var_12807, x = aw_chunk_763_cast)[name = tensor<string, []>("op_13220_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13221_cast = softmax(axis = var_12807, x = aw_chunk_765_cast)[name = tensor<string, []>("op_13221_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13222_cast = softmax(axis = var_12807, x = aw_chunk_767_cast)[name = tensor<string, []>("op_13222_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13223_cast = softmax(axis = var_12807, x = aw_chunk_769_cast)[name = tensor<string, []>("op_13223_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13224_cast = softmax(axis = var_12807, x = aw_chunk_771_cast)[name = tensor<string, []>("op_13224_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13225_cast = softmax(axis = var_12807, x = aw_chunk_773_cast)[name = tensor<string, []>("op_13225_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13226_cast = softmax(axis = var_12807, x = aw_chunk_775_cast)[name = tensor<string, []>("op_13226_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13227_cast = softmax(axis = var_12807, x = aw_chunk_777_cast)[name = tensor<string, []>("op_13227_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13228_cast = softmax(axis = var_12807, x = aw_chunk_779_cast)[name = tensor<string, []>("op_13228_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13229_cast = softmax(axis = var_12807, x = aw_chunk_781_cast)[name = tensor<string, []>("op_13229_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13230_cast = softmax(axis = var_12807, x = aw_chunk_783_cast)[name = tensor<string, []>("op_13230_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13231_cast = softmax(axis = var_12807, x = aw_chunk_785_cast)[name = tensor<string, []>("op_13231_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13232_cast = softmax(axis = var_12807, x = aw_chunk_787_cast)[name = tensor<string, []>("op_13232_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13233_cast = softmax(axis = var_12807, x = aw_chunk_789_cast)[name = tensor<string, []>("op_13233_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13234_cast = softmax(axis = var_12807, x = aw_chunk_791_cast)[name = tensor<string, []>("op_13234_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13235_cast = softmax(axis = var_12807, x = aw_chunk_793_cast)[name = tensor<string, []>("op_13235_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13236_cast = softmax(axis = var_12807, x = aw_chunk_795_cast)[name = tensor<string, []>("op_13236_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13237_cast = softmax(axis = var_12807, x = aw_chunk_797_cast)[name = tensor<string, []>("op_13237_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_13238_cast = softmax(axis = var_12807, x = aw_chunk_799_cast)[name = tensor<string, []>("op_13238_cast")];
            tensor<string, []> var_13240_equation_0 = const()[name = tensor<string, []>("op_13240_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13240_cast = einsum(equation = var_13240_equation_0, values = (var_13020_cast, var_13199_cast))[name = tensor<string, []>("op_13240_cast")];
            tensor<string, []> var_13242_equation_0 = const()[name = tensor<string, []>("op_13242_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13242_cast = einsum(equation = var_13242_equation_0, values = (var_13020_cast, var_13200_cast))[name = tensor<string, []>("op_13242_cast")];
            tensor<string, []> var_13244_equation_0 = const()[name = tensor<string, []>("op_13244_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13244_cast = einsum(equation = var_13244_equation_0, values = (var_13020_cast, var_13201_cast))[name = tensor<string, []>("op_13244_cast")];
            tensor<string, []> var_13246_equation_0 = const()[name = tensor<string, []>("op_13246_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13246_cast = einsum(equation = var_13246_equation_0, values = (var_13020_cast, var_13202_cast))[name = tensor<string, []>("op_13246_cast")];
            tensor<string, []> var_13248_equation_0 = const()[name = tensor<string, []>("op_13248_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13248_cast = einsum(equation = var_13248_equation_0, values = (var_13020_cast, var_13203_cast))[name = tensor<string, []>("op_13248_cast")];
            tensor<string, []> var_13250_equation_0 = const()[name = tensor<string, []>("op_13250_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13250_cast = einsum(equation = var_13250_equation_0, values = (var_13020_cast, var_13204_cast))[name = tensor<string, []>("op_13250_cast")];
            tensor<string, []> var_13252_equation_0 = const()[name = tensor<string, []>("op_13252_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13252_cast = einsum(equation = var_13252_equation_0, values = (var_13020_cast, var_13205_cast))[name = tensor<string, []>("op_13252_cast")];
            tensor<string, []> var_13254_equation_0 = const()[name = tensor<string, []>("op_13254_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13254_cast = einsum(equation = var_13254_equation_0, values = (var_13020_cast, var_13206_cast))[name = tensor<string, []>("op_13254_cast")];
            tensor<string, []> var_13256_equation_0 = const()[name = tensor<string, []>("op_13256_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13256_cast = einsum(equation = var_13256_equation_0, values = (var_13024_cast, var_13207_cast))[name = tensor<string, []>("op_13256_cast")];
            tensor<string, []> var_13258_equation_0 = const()[name = tensor<string, []>("op_13258_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13258_cast = einsum(equation = var_13258_equation_0, values = (var_13024_cast, var_13208_cast))[name = tensor<string, []>("op_13258_cast")];
            tensor<string, []> var_13260_equation_0 = const()[name = tensor<string, []>("op_13260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13260_cast = einsum(equation = var_13260_equation_0, values = (var_13024_cast, var_13209_cast))[name = tensor<string, []>("op_13260_cast")];
            tensor<string, []> var_13262_equation_0 = const()[name = tensor<string, []>("op_13262_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13262_cast = einsum(equation = var_13262_equation_0, values = (var_13024_cast, var_13210_cast))[name = tensor<string, []>("op_13262_cast")];
            tensor<string, []> var_13264_equation_0 = const()[name = tensor<string, []>("op_13264_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13264_cast = einsum(equation = var_13264_equation_0, values = (var_13024_cast, var_13211_cast))[name = tensor<string, []>("op_13264_cast")];
            tensor<string, []> var_13266_equation_0 = const()[name = tensor<string, []>("op_13266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13266_cast = einsum(equation = var_13266_equation_0, values = (var_13024_cast, var_13212_cast))[name = tensor<string, []>("op_13266_cast")];
            tensor<string, []> var_13268_equation_0 = const()[name = tensor<string, []>("op_13268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13268_cast = einsum(equation = var_13268_equation_0, values = (var_13024_cast, var_13213_cast))[name = tensor<string, []>("op_13268_cast")];
            tensor<string, []> var_13270_equation_0 = const()[name = tensor<string, []>("op_13270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13270_cast = einsum(equation = var_13270_equation_0, values = (var_13024_cast, var_13214_cast))[name = tensor<string, []>("op_13270_cast")];
            tensor<string, []> var_13272_equation_0 = const()[name = tensor<string, []>("op_13272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13272_cast = einsum(equation = var_13272_equation_0, values = (var_13028_cast, var_13215_cast))[name = tensor<string, []>("op_13272_cast")];
            tensor<string, []> var_13274_equation_0 = const()[name = tensor<string, []>("op_13274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13274_cast = einsum(equation = var_13274_equation_0, values = (var_13028_cast, var_13216_cast))[name = tensor<string, []>("op_13274_cast")];
            tensor<string, []> var_13276_equation_0 = const()[name = tensor<string, []>("op_13276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13276_cast = einsum(equation = var_13276_equation_0, values = (var_13028_cast, var_13217_cast))[name = tensor<string, []>("op_13276_cast")];
            tensor<string, []> var_13278_equation_0 = const()[name = tensor<string, []>("op_13278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13278_cast = einsum(equation = var_13278_equation_0, values = (var_13028_cast, var_13218_cast))[name = tensor<string, []>("op_13278_cast")];
            tensor<string, []> var_13280_equation_0 = const()[name = tensor<string, []>("op_13280_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13280_cast = einsum(equation = var_13280_equation_0, values = (var_13028_cast, var_13219_cast))[name = tensor<string, []>("op_13280_cast")];
            tensor<string, []> var_13282_equation_0 = const()[name = tensor<string, []>("op_13282_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13282_cast = einsum(equation = var_13282_equation_0, values = (var_13028_cast, var_13220_cast))[name = tensor<string, []>("op_13282_cast")];
            tensor<string, []> var_13284_equation_0 = const()[name = tensor<string, []>("op_13284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13284_cast = einsum(equation = var_13284_equation_0, values = (var_13028_cast, var_13221_cast))[name = tensor<string, []>("op_13284_cast")];
            tensor<string, []> var_13286_equation_0 = const()[name = tensor<string, []>("op_13286_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13286_cast = einsum(equation = var_13286_equation_0, values = (var_13028_cast, var_13222_cast))[name = tensor<string, []>("op_13286_cast")];
            tensor<string, []> var_13288_equation_0 = const()[name = tensor<string, []>("op_13288_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13288_cast = einsum(equation = var_13288_equation_0, values = (var_13032_cast, var_13223_cast))[name = tensor<string, []>("op_13288_cast")];
            tensor<string, []> var_13290_equation_0 = const()[name = tensor<string, []>("op_13290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13290_cast = einsum(equation = var_13290_equation_0, values = (var_13032_cast, var_13224_cast))[name = tensor<string, []>("op_13290_cast")];
            tensor<string, []> var_13292_equation_0 = const()[name = tensor<string, []>("op_13292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13292_cast = einsum(equation = var_13292_equation_0, values = (var_13032_cast, var_13225_cast))[name = tensor<string, []>("op_13292_cast")];
            tensor<string, []> var_13294_equation_0 = const()[name = tensor<string, []>("op_13294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13294_cast = einsum(equation = var_13294_equation_0, values = (var_13032_cast, var_13226_cast))[name = tensor<string, []>("op_13294_cast")];
            tensor<string, []> var_13296_equation_0 = const()[name = tensor<string, []>("op_13296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13296_cast = einsum(equation = var_13296_equation_0, values = (var_13032_cast, var_13227_cast))[name = tensor<string, []>("op_13296_cast")];
            tensor<string, []> var_13298_equation_0 = const()[name = tensor<string, []>("op_13298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13298_cast = einsum(equation = var_13298_equation_0, values = (var_13032_cast, var_13228_cast))[name = tensor<string, []>("op_13298_cast")];
            tensor<string, []> var_13300_equation_0 = const()[name = tensor<string, []>("op_13300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13300_cast = einsum(equation = var_13300_equation_0, values = (var_13032_cast, var_13229_cast))[name = tensor<string, []>("op_13300_cast")];
            tensor<string, []> var_13302_equation_0 = const()[name = tensor<string, []>("op_13302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13302_cast = einsum(equation = var_13302_equation_0, values = (var_13032_cast, var_13230_cast))[name = tensor<string, []>("op_13302_cast")];
            tensor<string, []> var_13304_equation_0 = const()[name = tensor<string, []>("op_13304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13304_cast = einsum(equation = var_13304_equation_0, values = (var_13036_cast, var_13231_cast))[name = tensor<string, []>("op_13304_cast")];
            tensor<string, []> var_13306_equation_0 = const()[name = tensor<string, []>("op_13306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13306_cast = einsum(equation = var_13306_equation_0, values = (var_13036_cast, var_13232_cast))[name = tensor<string, []>("op_13306_cast")];
            tensor<string, []> var_13308_equation_0 = const()[name = tensor<string, []>("op_13308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13308_cast = einsum(equation = var_13308_equation_0, values = (var_13036_cast, var_13233_cast))[name = tensor<string, []>("op_13308_cast")];
            tensor<string, []> var_13310_equation_0 = const()[name = tensor<string, []>("op_13310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13310_cast = einsum(equation = var_13310_equation_0, values = (var_13036_cast, var_13234_cast))[name = tensor<string, []>("op_13310_cast")];
            tensor<string, []> var_13312_equation_0 = const()[name = tensor<string, []>("op_13312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13312_cast = einsum(equation = var_13312_equation_0, values = (var_13036_cast, var_13235_cast))[name = tensor<string, []>("op_13312_cast")];
            tensor<string, []> var_13314_equation_0 = const()[name = tensor<string, []>("op_13314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13314_cast = einsum(equation = var_13314_equation_0, values = (var_13036_cast, var_13236_cast))[name = tensor<string, []>("op_13314_cast")];
            tensor<string, []> var_13316_equation_0 = const()[name = tensor<string, []>("op_13316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13316_cast = einsum(equation = var_13316_equation_0, values = (var_13036_cast, var_13237_cast))[name = tensor<string, []>("op_13316_cast")];
            tensor<string, []> var_13318_equation_0 = const()[name = tensor<string, []>("op_13318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13318_cast = einsum(equation = var_13318_equation_0, values = (var_13036_cast, var_13238_cast))[name = tensor<string, []>("op_13318_cast")];
            tensor<bool, []> var_13320_interleave_0 = const()[name = tensor<string, []>("op_13320_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13320_cast = concat(axis = var_12782, interleave = var_13320_interleave_0, values = (var_13240_cast, var_13242_cast, var_13244_cast, var_13246_cast, var_13248_cast, var_13250_cast, var_13252_cast, var_13254_cast))[name = tensor<string, []>("op_13320_cast")];
            tensor<bool, []> var_13322_interleave_0 = const()[name = tensor<string, []>("op_13322_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13322_cast = concat(axis = var_12782, interleave = var_13322_interleave_0, values = (var_13256_cast, var_13258_cast, var_13260_cast, var_13262_cast, var_13264_cast, var_13266_cast, var_13268_cast, var_13270_cast))[name = tensor<string, []>("op_13322_cast")];
            tensor<bool, []> var_13324_interleave_0 = const()[name = tensor<string, []>("op_13324_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13324_cast = concat(axis = var_12782, interleave = var_13324_interleave_0, values = (var_13272_cast, var_13274_cast, var_13276_cast, var_13278_cast, var_13280_cast, var_13282_cast, var_13284_cast, var_13286_cast))[name = tensor<string, []>("op_13324_cast")];
            tensor<bool, []> var_13326_interleave_0 = const()[name = tensor<string, []>("op_13326_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13326_cast = concat(axis = var_12782, interleave = var_13326_interleave_0, values = (var_13288_cast, var_13290_cast, var_13292_cast, var_13294_cast, var_13296_cast, var_13298_cast, var_13300_cast, var_13302_cast))[name = tensor<string, []>("op_13326_cast")];
            tensor<bool, []> var_13328_interleave_0 = const()[name = tensor<string, []>("op_13328_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13328_cast = concat(axis = var_12782, interleave = var_13328_interleave_0, values = (var_13304_cast, var_13306_cast, var_13308_cast, var_13310_cast, var_13312_cast, var_13314_cast, var_13316_cast, var_13318_cast))[name = tensor<string, []>("op_13328_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_12807, interleave = input_465_interleave_0, values = (var_13320_cast, var_13322_cast, var_13324_cast, var_13326_cast, var_13328_cast))[name = tensor<string, []>("input_465_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, 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213440192))), lut = tensor<fp16, [4]>([-0x1.a2p-4, -0x1.e8cp-6, 0x1.e5p-6, 0x1.a18p-4]), 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, []>(213465856)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13338_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_13336, groups = var_12807, pad = var_13338_pad_0, pad_type = var_13338_pad_type_0, strides = var_13334, 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_13338_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_81_cast = add(x = var_13338_cast, y = inputs_79_cast)[name = tensor<string, []>("inputs_81_cast")];
            tensor<int32, [1]> var_13342 = const()[name = tensor<string, []>("op_13342"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_81_cast = reduce_mean(axes = var_13342, keep_dims = var_12802, 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_13346 = const()[name = tensor<string, []>("op_13346"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_13347_cast = reduce_mean(axes = var_13346, keep_dims = var_12802, x = zero_mean_sq_81_cast)[name = tensor<string, []>("op_13347_cast")];
            tensor<fp16, []> var_13348_to_fp16 = const()[name = tensor<string, []>("op_13348_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_13349_cast = add(x = var_13347_cast, y = var_13348_to_fp16)[name = tensor<string, []>("op_13349_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_13349_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_13353_to_fp16 = const()[name = tensor<string, []>("op_13353_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213466560)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13354_cast = add(x = out_81_cast, y = var_13353_to_fp16)[name = tensor<string, []>("op_13354_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13356_to_fp16 = const()[name = tensor<string, []>("op_13356_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213467264)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_289_cast = mul(x = var_13354_cast, y = var_13356_to_fp16)[name = tensor<string, []>("hidden_states_289_cast")];
            tensor<int32, [2]> var_13363 = const()[name = tensor<string, []>("op_13363"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13365 = const()[name = tensor<string, []>("op_13365"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213467968))), lut = tensor<fp16, [4]>([-0x1.a88p-4, -0x1.058p-5, 0x1.dd8p-6, 0x1.a08p-4]), 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_13365, groups = var_12807, pad = q_55_pad_0, pad_type = q_55_pad_type_0, strides = var_13363, 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_13369 = const()[name = tensor<string, []>("op_13369"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13371 = const()[name = tensor<string, []>("op_13371"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_109_pad_type_0 = const()[name = tensor<string, []>("k_109_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_109_pad_0 = const()[name = tensor<string, []>("k_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213493632))), lut = tensor<fp16, [4]>([-0x1.49p-4, -0x1.72p-6, 0x1.8c8p-6, 0x1.4f8p-4]), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_109_cast = conv(dilations = var_13371, groups = var_12807, pad = k_109_pad_0, pad_type = k_109_pad_type_0, strides = var_13369, 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_13375 = const()[name = tensor<string, []>("op_13375"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13377 = const()[name = tensor<string, []>("op_13377"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_55_pad_type_0 = const()[name = tensor<string, []>("v_55_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_55_pad_0 = const()[name = tensor<string, []>("v_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213575616))), lut = tensor<fp16, [4]>([-0x1.fbcp-6, -0x1.22p-7, 0x1.264p-7, 0x1.008p-5]), name = tensor<string, []>("up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_55_cast = conv(dilations = var_13377, groups = var_12807, pad = v_55_pad_0, pad_type = v_55_pad_type_0, strides = var_13375, 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_13381_begin_0 = const()[name = tensor<string, []>("op_13381_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13381_end_0 = const()[name = tensor<string, []>("op_13381_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13381_end_mask_0 = const()[name = tensor<string, []>("op_13381_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13381_cast = slice_by_index(begin = var_13381_begin_0, end = var_13381_end_0, end_mask = var_13381_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_13381_cast")];
            tensor<int32, [4]> var_13385_begin_0 = const()[name = tensor<string, []>("op_13385_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13385_end_0 = const()[name = tensor<string, []>("op_13385_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_13385_end_mask_0 = const()[name = tensor<string, []>("op_13385_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13385_cast = slice_by_index(begin = var_13385_begin_0, end = var_13385_end_0, end_mask = var_13385_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_13385_cast")];
            tensor<int32, [4]> var_13389_begin_0 = const()[name = tensor<string, []>("op_13389_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13389_end_0 = const()[name = tensor<string, []>("op_13389_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_13389_end_mask_0 = const()[name = tensor<string, []>("op_13389_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13389_cast = slice_by_index(begin = var_13389_begin_0, end = var_13389_end_0, end_mask = var_13389_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_13389_cast")];
            tensor<int32, [4]> var_13393_begin_0 = const()[name = tensor<string, []>("op_13393_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13393_end_0 = const()[name = tensor<string, []>("op_13393_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_13393_end_mask_0 = const()[name = tensor<string, []>("op_13393_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13393_cast = slice_by_index(begin = var_13393_begin_0, end = var_13393_end_0, end_mask = var_13393_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_13393_cast")];
            tensor<int32, [4]> var_13397_begin_0 = const()[name = tensor<string, []>("op_13397_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13397_end_0 = const()[name = tensor<string, []>("op_13397_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_13397_end_mask_0 = const()[name = tensor<string, []>("op_13397_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13397_cast = slice_by_index(begin = var_13397_begin_0, end = var_13397_end_0, end_mask = var_13397_end_mask_0, x = q_55_cast)[name = tensor<string, []>("op_13397_cast")];
            tensor<int32, [4]> var_13400_begin_0 = const()[name = tensor<string, []>("op_13400_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13400_end_0 = const()[name = tensor<string, []>("op_13400_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13400_end_mask_0 = const()[name = tensor<string, []>("op_13400_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13400_cast = slice_by_index(begin = var_13400_begin_0, end = var_13400_end_0, end_mask = var_13400_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13400_cast")];
            tensor<int32, [4]> var_13401_begin_0 = const()[name = tensor<string, []>("op_13401_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13401_end_0 = const()[name = tensor<string, []>("op_13401_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13401_end_mask_0 = const()[name = tensor<string, []>("op_13401_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13401_cast = slice_by_index(begin = var_13401_begin_0, end = var_13401_end_0, end_mask = var_13401_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13401_cast")];
            tensor<int32, [4]> var_13402_begin_0 = const()[name = tensor<string, []>("op_13402_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13402_end_0 = const()[name = tensor<string, []>("op_13402_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13402_end_mask_0 = const()[name = tensor<string, []>("op_13402_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13402_cast = slice_by_index(begin = var_13402_begin_0, end = var_13402_end_0, end_mask = var_13402_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13402_cast")];
            tensor<int32, [4]> var_13403_begin_0 = const()[name = tensor<string, []>("op_13403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13403_end_0 = const()[name = tensor<string, []>("op_13403_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13403_end_mask_0 = const()[name = tensor<string, []>("op_13403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13403_cast = slice_by_index(begin = var_13403_begin_0, end = var_13403_end_0, end_mask = var_13403_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13403_cast")];
            tensor<int32, [4]> var_13404_begin_0 = const()[name = tensor<string, []>("op_13404_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13404_end_0 = const()[name = tensor<string, []>("op_13404_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13404_end_mask_0 = const()[name = tensor<string, []>("op_13404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13404_cast = slice_by_index(begin = var_13404_begin_0, end = var_13404_end_0, end_mask = var_13404_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13404_cast")];
            tensor<int32, [4]> var_13405_begin_0 = const()[name = tensor<string, []>("op_13405_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13405_end_0 = const()[name = tensor<string, []>("op_13405_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13405_end_mask_0 = const()[name = tensor<string, []>("op_13405_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13405_cast = slice_by_index(begin = var_13405_begin_0, end = var_13405_end_0, end_mask = var_13405_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13405_cast")];
            tensor<int32, [4]> var_13406_begin_0 = const()[name = tensor<string, []>("op_13406_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13406_end_0 = const()[name = tensor<string, []>("op_13406_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13406_end_mask_0 = const()[name = tensor<string, []>("op_13406_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13406_cast = slice_by_index(begin = var_13406_begin_0, end = var_13406_end_0, end_mask = var_13406_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13406_cast")];
            tensor<int32, [4]> var_13407_begin_0 = const()[name = tensor<string, []>("op_13407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13407_end_0 = const()[name = tensor<string, []>("op_13407_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13407_end_mask_0 = const()[name = tensor<string, []>("op_13407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13407_cast = slice_by_index(begin = var_13407_begin_0, end = var_13407_end_0, end_mask = var_13407_end_mask_0, x = var_13381_cast)[name = tensor<string, []>("op_13407_cast")];
            tensor<int32, [4]> var_13408_begin_0 = const()[name = tensor<string, []>("op_13408_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13408_end_0 = const()[name = tensor<string, []>("op_13408_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13408_end_mask_0 = const()[name = tensor<string, []>("op_13408_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13408_cast = slice_by_index(begin = var_13408_begin_0, end = var_13408_end_0, end_mask = var_13408_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13408_cast")];
            tensor<int32, [4]> var_13409_begin_0 = const()[name = tensor<string, []>("op_13409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13409_end_0 = const()[name = tensor<string, []>("op_13409_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13409_end_mask_0 = const()[name = tensor<string, []>("op_13409_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13409_cast = slice_by_index(begin = var_13409_begin_0, end = var_13409_end_0, end_mask = var_13409_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13409_cast")];
            tensor<int32, [4]> var_13410_begin_0 = const()[name = tensor<string, []>("op_13410_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13410_end_0 = const()[name = tensor<string, []>("op_13410_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13410_end_mask_0 = const()[name = tensor<string, []>("op_13410_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13410_cast = slice_by_index(begin = var_13410_begin_0, end = var_13410_end_0, end_mask = var_13410_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13410_cast")];
            tensor<int32, [4]> var_13411_begin_0 = const()[name = tensor<string, []>("op_13411_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13411_end_0 = const()[name = tensor<string, []>("op_13411_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13411_end_mask_0 = const()[name = tensor<string, []>("op_13411_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13411_cast = slice_by_index(begin = var_13411_begin_0, end = var_13411_end_0, end_mask = var_13411_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13411_cast")];
            tensor<int32, [4]> var_13412_begin_0 = const()[name = tensor<string, []>("op_13412_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13412_end_0 = const()[name = tensor<string, []>("op_13412_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13412_end_mask_0 = const()[name = tensor<string, []>("op_13412_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13412_cast = slice_by_index(begin = var_13412_begin_0, end = var_13412_end_0, end_mask = var_13412_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13412_cast")];
            tensor<int32, [4]> var_13413_begin_0 = const()[name = tensor<string, []>("op_13413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13413_end_0 = const()[name = tensor<string, []>("op_13413_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13413_end_mask_0 = const()[name = tensor<string, []>("op_13413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13413_cast = slice_by_index(begin = var_13413_begin_0, end = var_13413_end_0, end_mask = var_13413_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13413_cast")];
            tensor<int32, [4]> var_13414_begin_0 = const()[name = tensor<string, []>("op_13414_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13414_end_0 = const()[name = tensor<string, []>("op_13414_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13414_end_mask_0 = const()[name = tensor<string, []>("op_13414_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13414_cast = slice_by_index(begin = var_13414_begin_0, end = var_13414_end_0, end_mask = var_13414_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13414_cast")];
            tensor<int32, [4]> var_13415_begin_0 = const()[name = tensor<string, []>("op_13415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13415_end_0 = const()[name = tensor<string, []>("op_13415_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13415_end_mask_0 = const()[name = tensor<string, []>("op_13415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13415_cast = slice_by_index(begin = var_13415_begin_0, end = var_13415_end_0, end_mask = var_13415_end_mask_0, x = var_13385_cast)[name = tensor<string, []>("op_13415_cast")];
            tensor<int32, [4]> var_13416_begin_0 = const()[name = tensor<string, []>("op_13416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13416_end_0 = const()[name = tensor<string, []>("op_13416_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13416_end_mask_0 = const()[name = tensor<string, []>("op_13416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13416_cast = slice_by_index(begin = var_13416_begin_0, end = var_13416_end_0, end_mask = var_13416_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13416_cast")];
            tensor<int32, [4]> var_13417_begin_0 = const()[name = tensor<string, []>("op_13417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13417_end_0 = const()[name = tensor<string, []>("op_13417_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13417_end_mask_0 = const()[name = tensor<string, []>("op_13417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13417_cast = slice_by_index(begin = var_13417_begin_0, end = var_13417_end_0, end_mask = var_13417_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13417_cast")];
            tensor<int32, [4]> var_13418_begin_0 = const()[name = tensor<string, []>("op_13418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13418_end_0 = const()[name = tensor<string, []>("op_13418_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13418_end_mask_0 = const()[name = tensor<string, []>("op_13418_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13418_cast = slice_by_index(begin = var_13418_begin_0, end = var_13418_end_0, end_mask = var_13418_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13418_cast")];
            tensor<int32, [4]> var_13419_begin_0 = const()[name = tensor<string, []>("op_13419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13419_end_0 = const()[name = tensor<string, []>("op_13419_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13419_end_mask_0 = const()[name = tensor<string, []>("op_13419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13419_cast = slice_by_index(begin = var_13419_begin_0, end = var_13419_end_0, end_mask = var_13419_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13419_cast")];
            tensor<int32, [4]> var_13420_begin_0 = const()[name = tensor<string, []>("op_13420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13420_end_0 = const()[name = tensor<string, []>("op_13420_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13420_end_mask_0 = const()[name = tensor<string, []>("op_13420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13420_cast = slice_by_index(begin = var_13420_begin_0, end = var_13420_end_0, end_mask = var_13420_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13420_cast")];
            tensor<int32, [4]> var_13421_begin_0 = const()[name = tensor<string, []>("op_13421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13421_end_0 = const()[name = tensor<string, []>("op_13421_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13421_end_mask_0 = const()[name = tensor<string, []>("op_13421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13421_cast = slice_by_index(begin = var_13421_begin_0, end = var_13421_end_0, end_mask = var_13421_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13421_cast")];
            tensor<int32, [4]> var_13422_begin_0 = const()[name = tensor<string, []>("op_13422_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13422_end_0 = const()[name = tensor<string, []>("op_13422_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13422_end_mask_0 = const()[name = tensor<string, []>("op_13422_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13422_cast = slice_by_index(begin = var_13422_begin_0, end = var_13422_end_0, end_mask = var_13422_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13422_cast")];
            tensor<int32, [4]> var_13423_begin_0 = const()[name = tensor<string, []>("op_13423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13423_end_0 = const()[name = tensor<string, []>("op_13423_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13423_end_mask_0 = const()[name = tensor<string, []>("op_13423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13423_cast = slice_by_index(begin = var_13423_begin_0, end = var_13423_end_0, end_mask = var_13423_end_mask_0, x = var_13389_cast)[name = tensor<string, []>("op_13423_cast")];
            tensor<int32, [4]> var_13424_begin_0 = const()[name = tensor<string, []>("op_13424_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13424_end_0 = const()[name = tensor<string, []>("op_13424_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13424_end_mask_0 = const()[name = tensor<string, []>("op_13424_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13424_cast = slice_by_index(begin = var_13424_begin_0, end = var_13424_end_0, end_mask = var_13424_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13424_cast")];
            tensor<int32, [4]> var_13425_begin_0 = const()[name = tensor<string, []>("op_13425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13425_end_0 = const()[name = tensor<string, []>("op_13425_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13425_end_mask_0 = const()[name = tensor<string, []>("op_13425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13425_cast = slice_by_index(begin = var_13425_begin_0, end = var_13425_end_0, end_mask = var_13425_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13425_cast")];
            tensor<int32, [4]> var_13426_begin_0 = const()[name = tensor<string, []>("op_13426_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13426_end_0 = const()[name = tensor<string, []>("op_13426_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13426_end_mask_0 = const()[name = tensor<string, []>("op_13426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13426_cast = slice_by_index(begin = var_13426_begin_0, end = var_13426_end_0, end_mask = var_13426_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13426_cast")];
            tensor<int32, [4]> var_13427_begin_0 = const()[name = tensor<string, []>("op_13427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13427_end_0 = const()[name = tensor<string, []>("op_13427_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13427_end_mask_0 = const()[name = tensor<string, []>("op_13427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13427_cast = slice_by_index(begin = var_13427_begin_0, end = var_13427_end_0, end_mask = var_13427_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13427_cast")];
            tensor<int32, [4]> var_13428_begin_0 = const()[name = tensor<string, []>("op_13428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13428_end_0 = const()[name = tensor<string, []>("op_13428_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13428_end_mask_0 = const()[name = tensor<string, []>("op_13428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13428_cast = slice_by_index(begin = var_13428_begin_0, end = var_13428_end_0, end_mask = var_13428_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13428_cast")];
            tensor<int32, [4]> var_13429_begin_0 = const()[name = tensor<string, []>("op_13429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13429_end_0 = const()[name = tensor<string, []>("op_13429_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13429_end_mask_0 = const()[name = tensor<string, []>("op_13429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13429_cast = slice_by_index(begin = var_13429_begin_0, end = var_13429_end_0, end_mask = var_13429_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13429_cast")];
            tensor<int32, [4]> var_13430_begin_0 = const()[name = tensor<string, []>("op_13430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13430_end_0 = const()[name = tensor<string, []>("op_13430_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13430_end_mask_0 = const()[name = tensor<string, []>("op_13430_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13430_cast = slice_by_index(begin = var_13430_begin_0, end = var_13430_end_0, end_mask = var_13430_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13430_cast")];
            tensor<int32, [4]> var_13431_begin_0 = const()[name = tensor<string, []>("op_13431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13431_end_0 = const()[name = tensor<string, []>("op_13431_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13431_end_mask_0 = const()[name = tensor<string, []>("op_13431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13431_cast = slice_by_index(begin = var_13431_begin_0, end = var_13431_end_0, end_mask = var_13431_end_mask_0, x = var_13393_cast)[name = tensor<string, []>("op_13431_cast")];
            tensor<int32, [4]> var_13432_begin_0 = const()[name = tensor<string, []>("op_13432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13432_end_0 = const()[name = tensor<string, []>("op_13432_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13432_end_mask_0 = const()[name = tensor<string, []>("op_13432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13432_cast = slice_by_index(begin = var_13432_begin_0, end = var_13432_end_0, end_mask = var_13432_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13432_cast")];
            tensor<int32, [4]> var_13433_begin_0 = const()[name = tensor<string, []>("op_13433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13433_end_0 = const()[name = tensor<string, []>("op_13433_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13433_end_mask_0 = const()[name = tensor<string, []>("op_13433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13433_cast = slice_by_index(begin = var_13433_begin_0, end = var_13433_end_0, end_mask = var_13433_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13433_cast")];
            tensor<int32, [4]> var_13434_begin_0 = const()[name = tensor<string, []>("op_13434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13434_end_0 = const()[name = tensor<string, []>("op_13434_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13434_end_mask_0 = const()[name = tensor<string, []>("op_13434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13434_cast = slice_by_index(begin = var_13434_begin_0, end = var_13434_end_0, end_mask = var_13434_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13434_cast")];
            tensor<int32, [4]> var_13435_begin_0 = const()[name = tensor<string, []>("op_13435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13435_end_0 = const()[name = tensor<string, []>("op_13435_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13435_end_mask_0 = const()[name = tensor<string, []>("op_13435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13435_cast = slice_by_index(begin = var_13435_begin_0, end = var_13435_end_0, end_mask = var_13435_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13435_cast")];
            tensor<int32, [4]> var_13436_begin_0 = const()[name = tensor<string, []>("op_13436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13436_end_0 = const()[name = tensor<string, []>("op_13436_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13436_end_mask_0 = const()[name = tensor<string, []>("op_13436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13436_cast = slice_by_index(begin = var_13436_begin_0, end = var_13436_end_0, end_mask = var_13436_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13436_cast")];
            tensor<int32, [4]> var_13437_begin_0 = const()[name = tensor<string, []>("op_13437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13437_end_0 = const()[name = tensor<string, []>("op_13437_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13437_end_mask_0 = const()[name = tensor<string, []>("op_13437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13437_cast = slice_by_index(begin = var_13437_begin_0, end = var_13437_end_0, end_mask = var_13437_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13437_cast")];
            tensor<int32, [4]> var_13438_begin_0 = const()[name = tensor<string, []>("op_13438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13438_end_0 = const()[name = tensor<string, []>("op_13438_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13438_end_mask_0 = const()[name = tensor<string, []>("op_13438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13438_cast = slice_by_index(begin = var_13438_begin_0, end = var_13438_end_0, end_mask = var_13438_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13438_cast")];
            tensor<int32, [4]> var_13439_begin_0 = const()[name = tensor<string, []>("op_13439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13439_end_0 = const()[name = tensor<string, []>("op_13439_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13439_end_mask_0 = const()[name = tensor<string, []>("op_13439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13439_cast = slice_by_index(begin = var_13439_begin_0, end = var_13439_end_0, end_mask = var_13439_end_mask_0, x = var_13397_cast)[name = tensor<string, []>("op_13439_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_13444_begin_0 = const()[name = tensor<string, []>("op_13444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13444_end_0 = const()[name = tensor<string, []>("op_13444_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_13444_end_mask_0 = const()[name = tensor<string, []>("op_13444_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, 64]> var_13444_cast = slice_by_index(begin = var_13444_begin_0, end = var_13444_end_0, end_mask = var_13444_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_13444_cast")];
            tensor<int32, [4]> var_13448_begin_0 = const()[name = tensor<string, []>("op_13448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_13448_end_0 = const()[name = tensor<string, []>("op_13448_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_13448_end_mask_0 = const()[name = tensor<string, []>("op_13448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13448_cast = slice_by_index(begin = var_13448_begin_0, end = var_13448_end_0, end_mask = var_13448_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_13448_cast")];
            tensor<int32, [4]> var_13452_begin_0 = const()[name = tensor<string, []>("op_13452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_13452_end_0 = const()[name = tensor<string, []>("op_13452_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_13452_end_mask_0 = const()[name = tensor<string, []>("op_13452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13452_cast = slice_by_index(begin = var_13452_begin_0, end = var_13452_end_0, end_mask = var_13452_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_13452_cast")];
            tensor<int32, [4]> var_13456_begin_0 = const()[name = tensor<string, []>("op_13456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_13456_end_0 = const()[name = tensor<string, []>("op_13456_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_13456_end_mask_0 = const()[name = tensor<string, []>("op_13456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13456_cast = slice_by_index(begin = var_13456_begin_0, end = var_13456_end_0, end_mask = var_13456_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_13456_cast")];
            tensor<int32, [4]> var_13460_begin_0 = const()[name = tensor<string, []>("op_13460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_13460_end_0 = const()[name = tensor<string, []>("op_13460_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_13460_end_mask_0 = const()[name = tensor<string, []>("op_13460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_13460_cast = slice_by_index(begin = var_13460_begin_0, end = var_13460_end_0, end_mask = var_13460_end_mask_0, x = transpose_4)[name = tensor<string, []>("op_13460_cast")];
            tensor<int32, [4]> var_13462_begin_0 = const()[name = tensor<string, []>("op_13462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13462_end_0 = const()[name = tensor<string, []>("op_13462_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_13462_end_mask_0 = const()[name = tensor<string, []>("op_13462_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13462_cast = slice_by_index(begin = var_13462_begin_0, end = var_13462_end_0, end_mask = var_13462_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_13462_cast")];
            tensor<int32, [4]> var_13466_begin_0 = const()[name = tensor<string, []>("op_13466_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13466_end_0 = const()[name = tensor<string, []>("op_13466_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_13466_end_mask_0 = const()[name = tensor<string, []>("op_13466_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13466_cast = slice_by_index(begin = var_13466_begin_0, end = var_13466_end_0, end_mask = var_13466_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_13466_cast")];
            tensor<int32, [4]> var_13470_begin_0 = const()[name = tensor<string, []>("op_13470_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13470_end_0 = const()[name = tensor<string, []>("op_13470_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_13470_end_mask_0 = const()[name = tensor<string, []>("op_13470_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13470_cast = slice_by_index(begin = var_13470_begin_0, end = var_13470_end_0, end_mask = var_13470_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_13470_cast")];
            tensor<int32, [4]> var_13474_begin_0 = const()[name = tensor<string, []>("op_13474_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13474_end_0 = const()[name = tensor<string, []>("op_13474_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_13474_end_mask_0 = const()[name = tensor<string, []>("op_13474_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13474_cast = slice_by_index(begin = var_13474_begin_0, end = var_13474_end_0, end_mask = var_13474_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_13474_cast")];
            tensor<int32, [4]> var_13478_begin_0 = const()[name = tensor<string, []>("op_13478_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13478_end_0 = const()[name = tensor<string, []>("op_13478_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_13478_end_mask_0 = const()[name = tensor<string, []>("op_13478_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_13478_cast = slice_by_index(begin = var_13478_begin_0, end = var_13478_end_0, end_mask = var_13478_end_mask_0, x = v_55_cast)[name = tensor<string, []>("op_13478_cast")];
            tensor<string, []> var_13482_equation_0 = const()[name = tensor<string, []>("op_13482_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13482_cast = einsum(equation = var_13482_equation_0, values = (var_13444_cast, var_13400_cast))[name = tensor<string, []>("op_13482_cast")];
            tensor<fp16, []> var_13483_to_fp16 = const()[name = tensor<string, []>("op_13483_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_801_cast = mul(x = var_13482_cast, y = var_13483_to_fp16)[name = tensor<string, []>("aw_chunk_801_cast")];
            tensor<string, []> var_13486_equation_0 = const()[name = tensor<string, []>("op_13486_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13486_cast = einsum(equation = var_13486_equation_0, values = (var_13444_cast, var_13401_cast))[name = tensor<string, []>("op_13486_cast")];
            tensor<fp16, []> var_13487_to_fp16 = const()[name = tensor<string, []>("op_13487_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_803_cast = mul(x = var_13486_cast, y = var_13487_to_fp16)[name = tensor<string, []>("aw_chunk_803_cast")];
            tensor<string, []> var_13490_equation_0 = const()[name = tensor<string, []>("op_13490_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13490_cast = einsum(equation = var_13490_equation_0, values = (var_13444_cast, var_13402_cast))[name = tensor<string, []>("op_13490_cast")];
            tensor<fp16, []> var_13491_to_fp16 = const()[name = tensor<string, []>("op_13491_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_805_cast = mul(x = var_13490_cast, y = var_13491_to_fp16)[name = tensor<string, []>("aw_chunk_805_cast")];
            tensor<string, []> var_13494_equation_0 = const()[name = tensor<string, []>("op_13494_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13494_cast = einsum(equation = var_13494_equation_0, values = (var_13444_cast, var_13403_cast))[name = tensor<string, []>("op_13494_cast")];
            tensor<fp16, []> var_13495_to_fp16 = const()[name = tensor<string, []>("op_13495_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_807_cast = mul(x = var_13494_cast, y = var_13495_to_fp16)[name = tensor<string, []>("aw_chunk_807_cast")];
            tensor<string, []> var_13498_equation_0 = const()[name = tensor<string, []>("op_13498_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13498_cast = einsum(equation = var_13498_equation_0, values = (var_13444_cast, var_13404_cast))[name = tensor<string, []>("op_13498_cast")];
            tensor<fp16, []> var_13499_to_fp16 = const()[name = tensor<string, []>("op_13499_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_809_cast = mul(x = var_13498_cast, y = var_13499_to_fp16)[name = tensor<string, []>("aw_chunk_809_cast")];
            tensor<string, []> var_13502_equation_0 = const()[name = tensor<string, []>("op_13502_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13502_cast = einsum(equation = var_13502_equation_0, values = (var_13444_cast, var_13405_cast))[name = tensor<string, []>("op_13502_cast")];
            tensor<fp16, []> var_13503_to_fp16 = const()[name = tensor<string, []>("op_13503_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_811_cast = mul(x = var_13502_cast, y = var_13503_to_fp16)[name = tensor<string, []>("aw_chunk_811_cast")];
            tensor<string, []> var_13506_equation_0 = const()[name = tensor<string, []>("op_13506_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13506_cast = einsum(equation = var_13506_equation_0, values = (var_13444_cast, var_13406_cast))[name = tensor<string, []>("op_13506_cast")];
            tensor<fp16, []> var_13507_to_fp16 = const()[name = tensor<string, []>("op_13507_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_813_cast = mul(x = var_13506_cast, y = var_13507_to_fp16)[name = tensor<string, []>("aw_chunk_813_cast")];
            tensor<string, []> var_13510_equation_0 = const()[name = tensor<string, []>("op_13510_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13510_cast = einsum(equation = var_13510_equation_0, values = (var_13444_cast, var_13407_cast))[name = tensor<string, []>("op_13510_cast")];
            tensor<fp16, []> var_13511_to_fp16 = const()[name = tensor<string, []>("op_13511_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_815_cast = mul(x = var_13510_cast, y = var_13511_to_fp16)[name = tensor<string, []>("aw_chunk_815_cast")];
            tensor<string, []> var_13514_equation_0 = const()[name = tensor<string, []>("op_13514_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13514_cast = einsum(equation = var_13514_equation_0, values = (var_13448_cast, var_13408_cast))[name = tensor<string, []>("op_13514_cast")];
            tensor<fp16, []> var_13515_to_fp16 = const()[name = tensor<string, []>("op_13515_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_817_cast = mul(x = var_13514_cast, y = var_13515_to_fp16)[name = tensor<string, []>("aw_chunk_817_cast")];
            tensor<string, []> var_13518_equation_0 = const()[name = tensor<string, []>("op_13518_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13518_cast = einsum(equation = var_13518_equation_0, values = (var_13448_cast, var_13409_cast))[name = tensor<string, []>("op_13518_cast")];
            tensor<fp16, []> var_13519_to_fp16 = const()[name = tensor<string, []>("op_13519_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_819_cast = mul(x = var_13518_cast, y = var_13519_to_fp16)[name = tensor<string, []>("aw_chunk_819_cast")];
            tensor<string, []> var_13522_equation_0 = const()[name = tensor<string, []>("op_13522_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13522_cast = einsum(equation = var_13522_equation_0, values = (var_13448_cast, var_13410_cast))[name = tensor<string, []>("op_13522_cast")];
            tensor<fp16, []> var_13523_to_fp16 = const()[name = tensor<string, []>("op_13523_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_821_cast = mul(x = var_13522_cast, y = var_13523_to_fp16)[name = tensor<string, []>("aw_chunk_821_cast")];
            tensor<string, []> var_13526_equation_0 = const()[name = tensor<string, []>("op_13526_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13526_cast = einsum(equation = var_13526_equation_0, values = (var_13448_cast, var_13411_cast))[name = tensor<string, []>("op_13526_cast")];
            tensor<fp16, []> var_13527_to_fp16 = const()[name = tensor<string, []>("op_13527_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_823_cast = mul(x = var_13526_cast, y = var_13527_to_fp16)[name = tensor<string, []>("aw_chunk_823_cast")];
            tensor<string, []> var_13530_equation_0 = const()[name = tensor<string, []>("op_13530_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13530_cast = einsum(equation = var_13530_equation_0, values = (var_13448_cast, var_13412_cast))[name = tensor<string, []>("op_13530_cast")];
            tensor<fp16, []> var_13531_to_fp16 = const()[name = tensor<string, []>("op_13531_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_825_cast = mul(x = var_13530_cast, y = var_13531_to_fp16)[name = tensor<string, []>("aw_chunk_825_cast")];
            tensor<string, []> var_13534_equation_0 = const()[name = tensor<string, []>("op_13534_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13534_cast = einsum(equation = var_13534_equation_0, values = (var_13448_cast, var_13413_cast))[name = tensor<string, []>("op_13534_cast")];
            tensor<fp16, []> var_13535_to_fp16 = const()[name = tensor<string, []>("op_13535_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_827_cast = mul(x = var_13534_cast, y = var_13535_to_fp16)[name = tensor<string, []>("aw_chunk_827_cast")];
            tensor<string, []> var_13538_equation_0 = const()[name = tensor<string, []>("op_13538_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13538_cast = einsum(equation = var_13538_equation_0, values = (var_13448_cast, var_13414_cast))[name = tensor<string, []>("op_13538_cast")];
            tensor<fp16, []> var_13539_to_fp16 = const()[name = tensor<string, []>("op_13539_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_829_cast = mul(x = var_13538_cast, y = var_13539_to_fp16)[name = tensor<string, []>("aw_chunk_829_cast")];
            tensor<string, []> var_13542_equation_0 = const()[name = tensor<string, []>("op_13542_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13542_cast = einsum(equation = var_13542_equation_0, values = (var_13448_cast, var_13415_cast))[name = tensor<string, []>("op_13542_cast")];
            tensor<fp16, []> var_13543_to_fp16 = const()[name = tensor<string, []>("op_13543_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_831_cast = mul(x = var_13542_cast, y = var_13543_to_fp16)[name = tensor<string, []>("aw_chunk_831_cast")];
            tensor<string, []> var_13546_equation_0 = const()[name = tensor<string, []>("op_13546_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13546_cast = einsum(equation = var_13546_equation_0, values = (var_13452_cast, var_13416_cast))[name = tensor<string, []>("op_13546_cast")];
            tensor<fp16, []> var_13547_to_fp16 = const()[name = tensor<string, []>("op_13547_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_833_cast = mul(x = var_13546_cast, y = var_13547_to_fp16)[name = tensor<string, []>("aw_chunk_833_cast")];
            tensor<string, []> var_13550_equation_0 = const()[name = tensor<string, []>("op_13550_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13550_cast = einsum(equation = var_13550_equation_0, values = (var_13452_cast, var_13417_cast))[name = tensor<string, []>("op_13550_cast")];
            tensor<fp16, []> var_13551_to_fp16 = const()[name = tensor<string, []>("op_13551_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_835_cast = mul(x = var_13550_cast, y = var_13551_to_fp16)[name = tensor<string, []>("aw_chunk_835_cast")];
            tensor<string, []> var_13554_equation_0 = const()[name = tensor<string, []>("op_13554_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13554_cast = einsum(equation = var_13554_equation_0, values = (var_13452_cast, var_13418_cast))[name = tensor<string, []>("op_13554_cast")];
            tensor<fp16, []> var_13555_to_fp16 = const()[name = tensor<string, []>("op_13555_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_837_cast = mul(x = var_13554_cast, y = var_13555_to_fp16)[name = tensor<string, []>("aw_chunk_837_cast")];
            tensor<string, []> var_13558_equation_0 = const()[name = tensor<string, []>("op_13558_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13558_cast = einsum(equation = var_13558_equation_0, values = (var_13452_cast, var_13419_cast))[name = tensor<string, []>("op_13558_cast")];
            tensor<fp16, []> var_13559_to_fp16 = const()[name = tensor<string, []>("op_13559_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_839_cast = mul(x = var_13558_cast, y = var_13559_to_fp16)[name = tensor<string, []>("aw_chunk_839_cast")];
            tensor<string, []> var_13562_equation_0 = const()[name = tensor<string, []>("op_13562_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13562_cast = einsum(equation = var_13562_equation_0, values = (var_13452_cast, var_13420_cast))[name = tensor<string, []>("op_13562_cast")];
            tensor<fp16, []> var_13563_to_fp16 = const()[name = tensor<string, []>("op_13563_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_841_cast = mul(x = var_13562_cast, y = var_13563_to_fp16)[name = tensor<string, []>("aw_chunk_841_cast")];
            tensor<string, []> var_13566_equation_0 = const()[name = tensor<string, []>("op_13566_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13566_cast = einsum(equation = var_13566_equation_0, values = (var_13452_cast, var_13421_cast))[name = tensor<string, []>("op_13566_cast")];
            tensor<fp16, []> var_13567_to_fp16 = const()[name = tensor<string, []>("op_13567_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_843_cast = mul(x = var_13566_cast, y = var_13567_to_fp16)[name = tensor<string, []>("aw_chunk_843_cast")];
            tensor<string, []> var_13570_equation_0 = const()[name = tensor<string, []>("op_13570_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13570_cast = einsum(equation = var_13570_equation_0, values = (var_13452_cast, var_13422_cast))[name = tensor<string, []>("op_13570_cast")];
            tensor<fp16, []> var_13571_to_fp16 = const()[name = tensor<string, []>("op_13571_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_845_cast = mul(x = var_13570_cast, y = var_13571_to_fp16)[name = tensor<string, []>("aw_chunk_845_cast")];
            tensor<string, []> var_13574_equation_0 = const()[name = tensor<string, []>("op_13574_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13574_cast = einsum(equation = var_13574_equation_0, values = (var_13452_cast, var_13423_cast))[name = tensor<string, []>("op_13574_cast")];
            tensor<fp16, []> var_13575_to_fp16 = const()[name = tensor<string, []>("op_13575_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_847_cast = mul(x = var_13574_cast, y = var_13575_to_fp16)[name = tensor<string, []>("aw_chunk_847_cast")];
            tensor<string, []> var_13578_equation_0 = const()[name = tensor<string, []>("op_13578_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13578_cast = einsum(equation = var_13578_equation_0, values = (var_13456_cast, var_13424_cast))[name = tensor<string, []>("op_13578_cast")];
            tensor<fp16, []> var_13579_to_fp16 = const()[name = tensor<string, []>("op_13579_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_849_cast = mul(x = var_13578_cast, y = var_13579_to_fp16)[name = tensor<string, []>("aw_chunk_849_cast")];
            tensor<string, []> var_13582_equation_0 = const()[name = tensor<string, []>("op_13582_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13582_cast = einsum(equation = var_13582_equation_0, values = (var_13456_cast, var_13425_cast))[name = tensor<string, []>("op_13582_cast")];
            tensor<fp16, []> var_13583_to_fp16 = const()[name = tensor<string, []>("op_13583_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_851_cast = mul(x = var_13582_cast, y = var_13583_to_fp16)[name = tensor<string, []>("aw_chunk_851_cast")];
            tensor<string, []> var_13586_equation_0 = const()[name = tensor<string, []>("op_13586_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13586_cast = einsum(equation = var_13586_equation_0, values = (var_13456_cast, var_13426_cast))[name = tensor<string, []>("op_13586_cast")];
            tensor<fp16, []> var_13587_to_fp16 = const()[name = tensor<string, []>("op_13587_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_853_cast = mul(x = var_13586_cast, y = var_13587_to_fp16)[name = tensor<string, []>("aw_chunk_853_cast")];
            tensor<string, []> var_13590_equation_0 = const()[name = tensor<string, []>("op_13590_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13590_cast = einsum(equation = var_13590_equation_0, values = (var_13456_cast, var_13427_cast))[name = tensor<string, []>("op_13590_cast")];
            tensor<fp16, []> var_13591_to_fp16 = const()[name = tensor<string, []>("op_13591_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_855_cast = mul(x = var_13590_cast, y = var_13591_to_fp16)[name = tensor<string, []>("aw_chunk_855_cast")];
            tensor<string, []> var_13594_equation_0 = const()[name = tensor<string, []>("op_13594_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13594_cast = einsum(equation = var_13594_equation_0, values = (var_13456_cast, var_13428_cast))[name = tensor<string, []>("op_13594_cast")];
            tensor<fp16, []> var_13595_to_fp16 = const()[name = tensor<string, []>("op_13595_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_857_cast = mul(x = var_13594_cast, y = var_13595_to_fp16)[name = tensor<string, []>("aw_chunk_857_cast")];
            tensor<string, []> var_13598_equation_0 = const()[name = tensor<string, []>("op_13598_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13598_cast = einsum(equation = var_13598_equation_0, values = (var_13456_cast, var_13429_cast))[name = tensor<string, []>("op_13598_cast")];
            tensor<fp16, []> var_13599_to_fp16 = const()[name = tensor<string, []>("op_13599_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_859_cast = mul(x = var_13598_cast, y = var_13599_to_fp16)[name = tensor<string, []>("aw_chunk_859_cast")];
            tensor<string, []> var_13602_equation_0 = const()[name = tensor<string, []>("op_13602_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13602_cast = einsum(equation = var_13602_equation_0, values = (var_13456_cast, var_13430_cast))[name = tensor<string, []>("op_13602_cast")];
            tensor<fp16, []> var_13603_to_fp16 = const()[name = tensor<string, []>("op_13603_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_861_cast = mul(x = var_13602_cast, y = var_13603_to_fp16)[name = tensor<string, []>("aw_chunk_861_cast")];
            tensor<string, []> var_13606_equation_0 = const()[name = tensor<string, []>("op_13606_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13606_cast = einsum(equation = var_13606_equation_0, values = (var_13456_cast, var_13431_cast))[name = tensor<string, []>("op_13606_cast")];
            tensor<fp16, []> var_13607_to_fp16 = const()[name = tensor<string, []>("op_13607_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_863_cast = mul(x = var_13606_cast, y = var_13607_to_fp16)[name = tensor<string, []>("aw_chunk_863_cast")];
            tensor<string, []> var_13610_equation_0 = const()[name = tensor<string, []>("op_13610_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13610_cast = einsum(equation = var_13610_equation_0, values = (var_13460_cast, var_13432_cast))[name = tensor<string, []>("op_13610_cast")];
            tensor<fp16, []> var_13611_to_fp16 = const()[name = tensor<string, []>("op_13611_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_865_cast = mul(x = var_13610_cast, y = var_13611_to_fp16)[name = tensor<string, []>("aw_chunk_865_cast")];
            tensor<string, []> var_13614_equation_0 = const()[name = tensor<string, []>("op_13614_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13614_cast = einsum(equation = var_13614_equation_0, values = (var_13460_cast, var_13433_cast))[name = tensor<string, []>("op_13614_cast")];
            tensor<fp16, []> var_13615_to_fp16 = const()[name = tensor<string, []>("op_13615_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_867_cast = mul(x = var_13614_cast, y = var_13615_to_fp16)[name = tensor<string, []>("aw_chunk_867_cast")];
            tensor<string, []> var_13618_equation_0 = const()[name = tensor<string, []>("op_13618_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13618_cast = einsum(equation = var_13618_equation_0, values = (var_13460_cast, var_13434_cast))[name = tensor<string, []>("op_13618_cast")];
            tensor<fp16, []> var_13619_to_fp16 = const()[name = tensor<string, []>("op_13619_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_869_cast = mul(x = var_13618_cast, y = var_13619_to_fp16)[name = tensor<string, []>("aw_chunk_869_cast")];
            tensor<string, []> var_13622_equation_0 = const()[name = tensor<string, []>("op_13622_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13622_cast = einsum(equation = var_13622_equation_0, values = (var_13460_cast, var_13435_cast))[name = tensor<string, []>("op_13622_cast")];
            tensor<fp16, []> var_13623_to_fp16 = const()[name = tensor<string, []>("op_13623_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_871_cast = mul(x = var_13622_cast, y = var_13623_to_fp16)[name = tensor<string, []>("aw_chunk_871_cast")];
            tensor<string, []> var_13626_equation_0 = const()[name = tensor<string, []>("op_13626_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13626_cast = einsum(equation = var_13626_equation_0, values = (var_13460_cast, var_13436_cast))[name = tensor<string, []>("op_13626_cast")];
            tensor<fp16, []> var_13627_to_fp16 = const()[name = tensor<string, []>("op_13627_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_873_cast = mul(x = var_13626_cast, y = var_13627_to_fp16)[name = tensor<string, []>("aw_chunk_873_cast")];
            tensor<string, []> var_13630_equation_0 = const()[name = tensor<string, []>("op_13630_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13630_cast = einsum(equation = var_13630_equation_0, values = (var_13460_cast, var_13437_cast))[name = tensor<string, []>("op_13630_cast")];
            tensor<fp16, []> var_13631_to_fp16 = const()[name = tensor<string, []>("op_13631_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_875_cast = mul(x = var_13630_cast, y = var_13631_to_fp16)[name = tensor<string, []>("aw_chunk_875_cast")];
            tensor<string, []> var_13634_equation_0 = const()[name = tensor<string, []>("op_13634_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13634_cast = einsum(equation = var_13634_equation_0, values = (var_13460_cast, var_13438_cast))[name = tensor<string, []>("op_13634_cast")];
            tensor<fp16, []> var_13635_to_fp16 = const()[name = tensor<string, []>("op_13635_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_877_cast = mul(x = var_13634_cast, y = var_13635_to_fp16)[name = tensor<string, []>("aw_chunk_877_cast")];
            tensor<string, []> var_13638_equation_0 = const()[name = tensor<string, []>("op_13638_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_13638_cast = einsum(equation = var_13638_equation_0, values = (var_13460_cast, var_13439_cast))[name = tensor<string, []>("op_13638_cast")];
            tensor<fp16, []> var_13639_to_fp16 = const()[name = tensor<string, []>("op_13639_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_879_cast = mul(x = var_13638_cast, y = var_13639_to_fp16)[name = tensor<string, []>("aw_chunk_879_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13641_cast = softmax(axis = var_12807, x = aw_chunk_801_cast)[name = tensor<string, []>("op_13641_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13642_cast = softmax(axis = var_12807, x = aw_chunk_803_cast)[name = tensor<string, []>("op_13642_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13643_cast = softmax(axis = var_12807, x = aw_chunk_805_cast)[name = tensor<string, []>("op_13643_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13644_cast = softmax(axis = var_12807, x = aw_chunk_807_cast)[name = tensor<string, []>("op_13644_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13645_cast = softmax(axis = var_12807, x = aw_chunk_809_cast)[name = tensor<string, []>("op_13645_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13646_cast = softmax(axis = var_12807, x = aw_chunk_811_cast)[name = tensor<string, []>("op_13646_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13647_cast = softmax(axis = var_12807, x = aw_chunk_813_cast)[name = tensor<string, []>("op_13647_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13648_cast = softmax(axis = var_12807, x = aw_chunk_815_cast)[name = tensor<string, []>("op_13648_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13649_cast = softmax(axis = var_12807, x = aw_chunk_817_cast)[name = tensor<string, []>("op_13649_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13650_cast = softmax(axis = var_12807, x = aw_chunk_819_cast)[name = tensor<string, []>("op_13650_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13651_cast = softmax(axis = var_12807, x = aw_chunk_821_cast)[name = tensor<string, []>("op_13651_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13652_cast = softmax(axis = var_12807, x = aw_chunk_823_cast)[name = tensor<string, []>("op_13652_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13653_cast = softmax(axis = var_12807, x = aw_chunk_825_cast)[name = tensor<string, []>("op_13653_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13654_cast = softmax(axis = var_12807, x = aw_chunk_827_cast)[name = tensor<string, []>("op_13654_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13655_cast = softmax(axis = var_12807, x = aw_chunk_829_cast)[name = tensor<string, []>("op_13655_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13656_cast = softmax(axis = var_12807, x = aw_chunk_831_cast)[name = tensor<string, []>("op_13656_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13657_cast = softmax(axis = var_12807, x = aw_chunk_833_cast)[name = tensor<string, []>("op_13657_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13658_cast = softmax(axis = var_12807, x = aw_chunk_835_cast)[name = tensor<string, []>("op_13658_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13659_cast = softmax(axis = var_12807, x = aw_chunk_837_cast)[name = tensor<string, []>("op_13659_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13660_cast = softmax(axis = var_12807, x = aw_chunk_839_cast)[name = tensor<string, []>("op_13660_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13661_cast = softmax(axis = var_12807, x = aw_chunk_841_cast)[name = tensor<string, []>("op_13661_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13662_cast = softmax(axis = var_12807, x = aw_chunk_843_cast)[name = tensor<string, []>("op_13662_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13663_cast = softmax(axis = var_12807, x = aw_chunk_845_cast)[name = tensor<string, []>("op_13663_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13664_cast = softmax(axis = var_12807, x = aw_chunk_847_cast)[name = tensor<string, []>("op_13664_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13665_cast = softmax(axis = var_12807, x = aw_chunk_849_cast)[name = tensor<string, []>("op_13665_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13666_cast = softmax(axis = var_12807, x = aw_chunk_851_cast)[name = tensor<string, []>("op_13666_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13667_cast = softmax(axis = var_12807, x = aw_chunk_853_cast)[name = tensor<string, []>("op_13667_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13668_cast = softmax(axis = var_12807, x = aw_chunk_855_cast)[name = tensor<string, []>("op_13668_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13669_cast = softmax(axis = var_12807, x = aw_chunk_857_cast)[name = tensor<string, []>("op_13669_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13670_cast = softmax(axis = var_12807, x = aw_chunk_859_cast)[name = tensor<string, []>("op_13670_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13671_cast = softmax(axis = var_12807, x = aw_chunk_861_cast)[name = tensor<string, []>("op_13671_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13672_cast = softmax(axis = var_12807, x = aw_chunk_863_cast)[name = tensor<string, []>("op_13672_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13673_cast = softmax(axis = var_12807, x = aw_chunk_865_cast)[name = tensor<string, []>("op_13673_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13674_cast = softmax(axis = var_12807, x = aw_chunk_867_cast)[name = tensor<string, []>("op_13674_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13675_cast = softmax(axis = var_12807, x = aw_chunk_869_cast)[name = tensor<string, []>("op_13675_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13676_cast = softmax(axis = var_12807, x = aw_chunk_871_cast)[name = tensor<string, []>("op_13676_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13677_cast = softmax(axis = var_12807, x = aw_chunk_873_cast)[name = tensor<string, []>("op_13677_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13678_cast = softmax(axis = var_12807, x = aw_chunk_875_cast)[name = tensor<string, []>("op_13678_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13679_cast = softmax(axis = var_12807, x = aw_chunk_877_cast)[name = tensor<string, []>("op_13679_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_13680_cast = softmax(axis = var_12807, x = aw_chunk_879_cast)[name = tensor<string, []>("op_13680_cast")];
            tensor<string, []> var_13682_equation_0 = const()[name = tensor<string, []>("op_13682_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13682_cast = einsum(equation = var_13682_equation_0, values = (var_13462_cast, var_13641_cast))[name = tensor<string, []>("op_13682_cast")];
            tensor<string, []> var_13684_equation_0 = const()[name = tensor<string, []>("op_13684_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13684_cast = einsum(equation = var_13684_equation_0, values = (var_13462_cast, var_13642_cast))[name = tensor<string, []>("op_13684_cast")];
            tensor<string, []> var_13686_equation_0 = const()[name = tensor<string, []>("op_13686_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13686_cast = einsum(equation = var_13686_equation_0, values = (var_13462_cast, var_13643_cast))[name = tensor<string, []>("op_13686_cast")];
            tensor<string, []> var_13688_equation_0 = const()[name = tensor<string, []>("op_13688_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13688_cast = einsum(equation = var_13688_equation_0, values = (var_13462_cast, var_13644_cast))[name = tensor<string, []>("op_13688_cast")];
            tensor<string, []> var_13690_equation_0 = const()[name = tensor<string, []>("op_13690_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13690_cast = einsum(equation = var_13690_equation_0, values = (var_13462_cast, var_13645_cast))[name = tensor<string, []>("op_13690_cast")];
            tensor<string, []> var_13692_equation_0 = const()[name = tensor<string, []>("op_13692_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13692_cast = einsum(equation = var_13692_equation_0, values = (var_13462_cast, var_13646_cast))[name = tensor<string, []>("op_13692_cast")];
            tensor<string, []> var_13694_equation_0 = const()[name = tensor<string, []>("op_13694_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13694_cast = einsum(equation = var_13694_equation_0, values = (var_13462_cast, var_13647_cast))[name = tensor<string, []>("op_13694_cast")];
            tensor<string, []> var_13696_equation_0 = const()[name = tensor<string, []>("op_13696_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13696_cast = einsum(equation = var_13696_equation_0, values = (var_13462_cast, var_13648_cast))[name = tensor<string, []>("op_13696_cast")];
            tensor<string, []> var_13698_equation_0 = const()[name = tensor<string, []>("op_13698_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13698_cast = einsum(equation = var_13698_equation_0, values = (var_13466_cast, var_13649_cast))[name = tensor<string, []>("op_13698_cast")];
            tensor<string, []> var_13700_equation_0 = const()[name = tensor<string, []>("op_13700_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13700_cast = einsum(equation = var_13700_equation_0, values = (var_13466_cast, var_13650_cast))[name = tensor<string, []>("op_13700_cast")];
            tensor<string, []> var_13702_equation_0 = const()[name = tensor<string, []>("op_13702_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13702_cast = einsum(equation = var_13702_equation_0, values = (var_13466_cast, var_13651_cast))[name = tensor<string, []>("op_13702_cast")];
            tensor<string, []> var_13704_equation_0 = const()[name = tensor<string, []>("op_13704_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13704_cast = einsum(equation = var_13704_equation_0, values = (var_13466_cast, var_13652_cast))[name = tensor<string, []>("op_13704_cast")];
            tensor<string, []> var_13706_equation_0 = const()[name = tensor<string, []>("op_13706_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13706_cast = einsum(equation = var_13706_equation_0, values = (var_13466_cast, var_13653_cast))[name = tensor<string, []>("op_13706_cast")];
            tensor<string, []> var_13708_equation_0 = const()[name = tensor<string, []>("op_13708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13708_cast = einsum(equation = var_13708_equation_0, values = (var_13466_cast, var_13654_cast))[name = tensor<string, []>("op_13708_cast")];
            tensor<string, []> var_13710_equation_0 = const()[name = tensor<string, []>("op_13710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13710_cast = einsum(equation = var_13710_equation_0, values = (var_13466_cast, var_13655_cast))[name = tensor<string, []>("op_13710_cast")];
            tensor<string, []> var_13712_equation_0 = const()[name = tensor<string, []>("op_13712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13712_cast = einsum(equation = var_13712_equation_0, values = (var_13466_cast, var_13656_cast))[name = tensor<string, []>("op_13712_cast")];
            tensor<string, []> var_13714_equation_0 = const()[name = tensor<string, []>("op_13714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13714_cast = einsum(equation = var_13714_equation_0, values = (var_13470_cast, var_13657_cast))[name = tensor<string, []>("op_13714_cast")];
            tensor<string, []> var_13716_equation_0 = const()[name = tensor<string, []>("op_13716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13716_cast = einsum(equation = var_13716_equation_0, values = (var_13470_cast, var_13658_cast))[name = tensor<string, []>("op_13716_cast")];
            tensor<string, []> var_13718_equation_0 = const()[name = tensor<string, []>("op_13718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13718_cast = einsum(equation = var_13718_equation_0, values = (var_13470_cast, var_13659_cast))[name = tensor<string, []>("op_13718_cast")];
            tensor<string, []> var_13720_equation_0 = const()[name = tensor<string, []>("op_13720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13720_cast = einsum(equation = var_13720_equation_0, values = (var_13470_cast, var_13660_cast))[name = tensor<string, []>("op_13720_cast")];
            tensor<string, []> var_13722_equation_0 = const()[name = tensor<string, []>("op_13722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13722_cast = einsum(equation = var_13722_equation_0, values = (var_13470_cast, var_13661_cast))[name = tensor<string, []>("op_13722_cast")];
            tensor<string, []> var_13724_equation_0 = const()[name = tensor<string, []>("op_13724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13724_cast = einsum(equation = var_13724_equation_0, values = (var_13470_cast, var_13662_cast))[name = tensor<string, []>("op_13724_cast")];
            tensor<string, []> var_13726_equation_0 = const()[name = tensor<string, []>("op_13726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13726_cast = einsum(equation = var_13726_equation_0, values = (var_13470_cast, var_13663_cast))[name = tensor<string, []>("op_13726_cast")];
            tensor<string, []> var_13728_equation_0 = const()[name = tensor<string, []>("op_13728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13728_cast = einsum(equation = var_13728_equation_0, values = (var_13470_cast, var_13664_cast))[name = tensor<string, []>("op_13728_cast")];
            tensor<string, []> var_13730_equation_0 = const()[name = tensor<string, []>("op_13730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13730_cast = einsum(equation = var_13730_equation_0, values = (var_13474_cast, var_13665_cast))[name = tensor<string, []>("op_13730_cast")];
            tensor<string, []> var_13732_equation_0 = const()[name = tensor<string, []>("op_13732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13732_cast = einsum(equation = var_13732_equation_0, values = (var_13474_cast, var_13666_cast))[name = tensor<string, []>("op_13732_cast")];
            tensor<string, []> var_13734_equation_0 = const()[name = tensor<string, []>("op_13734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13734_cast = einsum(equation = var_13734_equation_0, values = (var_13474_cast, var_13667_cast))[name = tensor<string, []>("op_13734_cast")];
            tensor<string, []> var_13736_equation_0 = const()[name = tensor<string, []>("op_13736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13736_cast = einsum(equation = var_13736_equation_0, values = (var_13474_cast, var_13668_cast))[name = tensor<string, []>("op_13736_cast")];
            tensor<string, []> var_13738_equation_0 = const()[name = tensor<string, []>("op_13738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13738_cast = einsum(equation = var_13738_equation_0, values = (var_13474_cast, var_13669_cast))[name = tensor<string, []>("op_13738_cast")];
            tensor<string, []> var_13740_equation_0 = const()[name = tensor<string, []>("op_13740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13740_cast = einsum(equation = var_13740_equation_0, values = (var_13474_cast, var_13670_cast))[name = tensor<string, []>("op_13740_cast")];
            tensor<string, []> var_13742_equation_0 = const()[name = tensor<string, []>("op_13742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13742_cast = einsum(equation = var_13742_equation_0, values = (var_13474_cast, var_13671_cast))[name = tensor<string, []>("op_13742_cast")];
            tensor<string, []> var_13744_equation_0 = const()[name = tensor<string, []>("op_13744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13744_cast = einsum(equation = var_13744_equation_0, values = (var_13474_cast, var_13672_cast))[name = tensor<string, []>("op_13744_cast")];
            tensor<string, []> var_13746_equation_0 = const()[name = tensor<string, []>("op_13746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13746_cast = einsum(equation = var_13746_equation_0, values = (var_13478_cast, var_13673_cast))[name = tensor<string, []>("op_13746_cast")];
            tensor<string, []> var_13748_equation_0 = const()[name = tensor<string, []>("op_13748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13748_cast = einsum(equation = var_13748_equation_0, values = (var_13478_cast, var_13674_cast))[name = tensor<string, []>("op_13748_cast")];
            tensor<string, []> var_13750_equation_0 = const()[name = tensor<string, []>("op_13750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13750_cast = einsum(equation = var_13750_equation_0, values = (var_13478_cast, var_13675_cast))[name = tensor<string, []>("op_13750_cast")];
            tensor<string, []> var_13752_equation_0 = const()[name = tensor<string, []>("op_13752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13752_cast = einsum(equation = var_13752_equation_0, values = (var_13478_cast, var_13676_cast))[name = tensor<string, []>("op_13752_cast")];
            tensor<string, []> var_13754_equation_0 = const()[name = tensor<string, []>("op_13754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13754_cast = einsum(equation = var_13754_equation_0, values = (var_13478_cast, var_13677_cast))[name = tensor<string, []>("op_13754_cast")];
            tensor<string, []> var_13756_equation_0 = const()[name = tensor<string, []>("op_13756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13756_cast = einsum(equation = var_13756_equation_0, values = (var_13478_cast, var_13678_cast))[name = tensor<string, []>("op_13756_cast")];
            tensor<string, []> var_13758_equation_0 = const()[name = tensor<string, []>("op_13758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13758_cast = einsum(equation = var_13758_equation_0, values = (var_13478_cast, var_13679_cast))[name = tensor<string, []>("op_13758_cast")];
            tensor<string, []> var_13760_equation_0 = const()[name = tensor<string, []>("op_13760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_13760_cast = einsum(equation = var_13760_equation_0, values = (var_13478_cast, var_13680_cast))[name = tensor<string, []>("op_13760_cast")];
            tensor<bool, []> var_13762_interleave_0 = const()[name = tensor<string, []>("op_13762_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13762_cast = concat(axis = var_12782, interleave = var_13762_interleave_0, values = (var_13682_cast, var_13684_cast, var_13686_cast, var_13688_cast, var_13690_cast, var_13692_cast, var_13694_cast, var_13696_cast))[name = tensor<string, []>("op_13762_cast")];
            tensor<bool, []> var_13764_interleave_0 = const()[name = tensor<string, []>("op_13764_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13764_cast = concat(axis = var_12782, interleave = var_13764_interleave_0, values = (var_13698_cast, var_13700_cast, var_13702_cast, var_13704_cast, var_13706_cast, var_13708_cast, var_13710_cast, var_13712_cast))[name = tensor<string, []>("op_13764_cast")];
            tensor<bool, []> var_13766_interleave_0 = const()[name = tensor<string, []>("op_13766_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13766_cast = concat(axis = var_12782, interleave = var_13766_interleave_0, values = (var_13714_cast, var_13716_cast, var_13718_cast, var_13720_cast, var_13722_cast, var_13724_cast, var_13726_cast, var_13728_cast))[name = tensor<string, []>("op_13766_cast")];
            tensor<bool, []> var_13768_interleave_0 = const()[name = tensor<string, []>("op_13768_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13768_cast = concat(axis = var_12782, interleave = var_13768_interleave_0, values = (var_13730_cast, var_13732_cast, var_13734_cast, var_13736_cast, var_13738_cast, var_13740_cast, var_13742_cast, var_13744_cast))[name = tensor<string, []>("op_13768_cast")];
            tensor<bool, []> var_13770_interleave_0 = const()[name = tensor<string, []>("op_13770_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_13770_cast = concat(axis = var_12782, interleave = var_13770_interleave_0, values = (var_13746_cast, var_13748_cast, var_13750_cast, var_13752_cast, var_13754_cast, var_13756_cast, var_13758_cast, var_13760_cast))[name = tensor<string, []>("op_13770_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_12807, interleave = input_467_interleave_0, values = (var_13762_cast, var_13764_cast, var_13766_cast, var_13768_cast, var_13770_cast))[name = tensor<string, []>("input_467_cast")];
            tensor<int32, [2]> var_13776 = const()[name = tensor<string, []>("op_13776"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13778 = const()[name = tensor<string, []>("op_13778"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13780_pad_type_0 = const()[name = tensor<string, []>("op_13780_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13780_pad_0 = const()[name = tensor<string, []>("op_13780_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213657600))), lut = tensor<fp16, [4]>([-0x1.de8p-6, -0x1.d18p-9, 0x1.e4cp-9, 0x1.e78p-6]), 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, []>(213683264)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13780_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_13778, groups = var_12807, pad = var_13780_pad_0, pad_type = var_13780_pad_type_0, strides = var_13776, 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_13780_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_83_cast = add(x = var_13780_cast, y = inputs_81_cast)[name = tensor<string, []>("inputs_83_cast")];
            tensor<int32, [1]> var_13784 = const()[name = tensor<string, []>("op_13784"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_83_cast = reduce_mean(axes = var_13784, keep_dims = var_12802, 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_13788 = const()[name = tensor<string, []>("op_13788"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_13789_cast = reduce_mean(axes = var_13788, keep_dims = var_12802, x = zero_mean_sq_83_cast)[name = tensor<string, []>("op_13789_cast")];
            tensor<fp16, []> var_13790_to_fp16 = const()[name = tensor<string, []>("op_13790_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_13791_cast = add(x = var_13789_cast, y = var_13790_to_fp16)[name = tensor<string, []>("op_13791_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_13791_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_13795_to_fp16 = const()[name = tensor<string, []>("op_13795_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213683968)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13796_cast = add(x = out_83_cast, y = var_13795_to_fp16)[name = tensor<string, []>("op_13796_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13798_to_fp16 = const()[name = tensor<string, []>("op_13798_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213684672)))];
            tensor<fp16, [2, 320, 1, 4096]> input_469_cast = mul(x = var_13796_cast, y = var_13798_to_fp16)[name = tensor<string, []>("input_469_cast")];
            tensor<int32, [2]> var_13806 = const()[name = tensor<string, []>("op_13806"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13808 = const()[name = tensor<string, []>("op_13808"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13810_pad_type_0 = const()[name = tensor<string, []>("op_13810_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13810_pad_0 = const()[name = tensor<string, []>("op_13810_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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213685376))), lut = tensor<fp16, [4]>([-0x1.7dcp-4, -0x1.c1cp-6, 0x1.bfcp-6, 0x1.7e4p-4]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213890240))), lut = tensor<fp16, [4]>([0x1.1a8p-5, 0x1.9c4p-4, -0x1.3d4p-4, -0x1.f38p-7]), 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_13810_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_13808, groups = var_12807, pad = var_13810_pad_0, pad_type = var_13810_pad_type_0, strides = var_13806, 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_13810_cast")];
            tensor<int32, [2]> var_13811_split_sizes_0 = const()[name = tensor<string, []>("op_13811_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_13811_axis_0 = const()[name = tensor<string, []>("op_13811_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_13811_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_13811_cast_1 = split(axis = var_13811_axis_0, split_sizes = var_13811_split_sizes_0, x = var_13810_cast)[name = tensor<string, []>("op_13811_cast")];
            tensor<string, []> var_13813_mode_0 = const()[name = tensor<string, []>("op_13813_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_13813_cast = gelu(mode = var_13813_mode_0, x = var_13811_cast_1)[name = tensor<string, []>("op_13813_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_471_cast = mul(x = var_13811_cast_0, y = var_13813_cast)[name = tensor<string, []>("input_471_cast")];
            tensor<int32, [2]> var_13817 = const()[name = tensor<string, []>("op_13817"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13819 = const()[name = tensor<string, []>("op_13819"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_13821_pad_type_0 = const()[name = tensor<string, []>("op_13821_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_13821_pad_0 = const()[name = tensor<string, []>("op_13821_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213890944))), lut = tensor<fp16, [4]>([-0x1.9dcp-4, -0x1.e78p-6, 0x1.ee8p-6, 0x1.ap-4]), 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, []>(213993408)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13821_cast = conv(bias = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_13819, groups = var_12807, pad = var_13821_pad_0, pad_type = var_13821_pad_type_0, strides = var_13817, weight = up_blocks_3_attentions_0_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_471_cast)[name = tensor<string, []>("op_13821_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_293_cast = add(x = var_13821_cast, y = inputs_83_cast)[name = tensor<string, []>("hidden_states_293_cast")];
            tensor<int32, [4]> var_13823 = const()[name = tensor<string, []>("op_13823"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_473_cast = reshape(shape = var_13823, x = hidden_states_293_cast)[name = tensor<string, []>("input_473_cast")];
            tensor<int32, [2]> var_13827 = const()[name = tensor<string, []>("op_13827"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13829 = const()[name = tensor<string, []>("op_13829"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213994112))), lut = tensor<fp16, [4]>([-0x1.378p-3, -0x1.6dp-5, 0x1.774p-5, 0x1.3c4p-3]), 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, []>(214019776)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_295_cast = conv(bias = up_blocks_3_attentions_0_proj_out_bias_to_fp16, dilations = var_13829, groups = var_12807, pad = hidden_states_295_pad_0, pad_type = hidden_states_295_pad_type_0, strides = var_13827, 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_12807, 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, []>(214020480)))];
            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, []>(214021824)))];
            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_13847 = const()[name = tensor<string, []>("op_13847"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13849 = const()[name = tensor<string, []>("op_13849"), 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, [460800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214023168))), lut = tensor<fp16, [4]>([-0x1.a4cp-4, -0x1.7fcp-6, 0x1.b3cp-6, 0x1.bb4p-4]), 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, []>(214484032)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_299_cast = conv(bias = up_blocks_3_resnets_1_conv1_bias_to_fp16, dilations = var_13849, groups = var_12807, pad = hidden_states_299_pad_0, pad_type = hidden_states_299_pad_type_0, strides = var_13847, 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_13855 = const()[name = tensor<string, []>("op_13855"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13857 = const()[name = tensor<string, []>("op_13857"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214484736))), lut = tensor<fp16, [4]>([-0x1.598p-5, -0x1.7p-9, 0x1.608p-9, 0x1.3acp-5]), 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, []>(214587200)))];
            tensor<fp16, [2, 320, 1, 1]> temb_41_cast = conv(bias = up_blocks_3_resnets_1_time_emb_proj_bias_to_fp16, dilations = var_13857, groups = var_12807, pad = temb_41_pad_0, pad_type = temb_41_pad_type_0, strides = var_13855, 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, []>(214587904)))];
            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, []>(214588608)))];
            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_13867 = const()[name = tensor<string, []>("op_13867"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13869 = const()[name = tensor<string, []>("op_13869"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214589312))), lut = tensor<fp16, [4]>([-0x1.b68p-4, -0x1.93p-6, 0x1.7ep-6, 0x1.ac4p-4]), 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, []>(214819776)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_301_cast = conv(bias = up_blocks_3_resnets_1_conv2_bias_to_fp16, dilations = var_13869, groups = var_12807, pad = hidden_states_301_pad_0, pad_type = hidden_states_301_pad_type_0, strides = var_13867, 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_13874 = const()[name = tensor<string, []>("op_13874"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13876 = const()[name = tensor<string, []>("op_13876"), 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, [51200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214820480))), lut = tensor<fp16, [4]>([-0x1.a54p-5, -0x1.91cp-7, 0x1.cep-7, 0x1.becp-5]), 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, []>(214871744)))];
            tensor<fp16, [2, 320, 64, 64]> x_25_cast = conv(bias = up_blocks_3_resnets_1_conv_shortcut_bias_to_fp16, dilations = var_13876, groups = var_12807, pad = x_25_pad_0, pad_type = x_25_pad_type_0, strides = var_13874, 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, []>(214872448)))];
            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, []>(214873152)))];
            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_13896 = const()[name = tensor<string, []>("op_13896"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13898 = const()[name = tensor<string, []>("op_13898"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214873856))), lut = tensor<fp16, [4]>([-0x1.9d8p-4, -0x1.d9cp-6, 0x1.edp-6, 0x1.a3p-4]), 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, []>(214899520)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_305_cast = conv(bias = up_blocks_3_attentions_1_proj_in_bias_to_fp16, dilations = var_13898, groups = var_12807, pad = hidden_states_305_pad_0, pad_type = hidden_states_305_pad_type_0, strides = var_13896, 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_13903 = const()[name = tensor<string, []>("op_13903"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_85_cast = reshape(shape = var_13903, x = hidden_states_305_cast)[name = tensor<string, []>("inputs_85_cast")];
            tensor<int32, [1]> var_13913 = const()[name = tensor<string, []>("op_13913"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_85_cast = reduce_mean(axes = var_13913, keep_dims = var_12802, 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_13917 = const()[name = tensor<string, []>("op_13917"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_13918_cast = reduce_mean(axes = var_13917, keep_dims = var_12802, x = zero_mean_sq_85_cast)[name = tensor<string, []>("op_13918_cast")];
            tensor<fp16, []> var_13919_to_fp16 = const()[name = tensor<string, []>("op_13919_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_13920_cast = add(x = var_13918_cast, y = var_13919_to_fp16)[name = tensor<string, []>("op_13920_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_13920_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_13924_to_fp16 = const()[name = tensor<string, []>("op_13924_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214900224)))];
            tensor<fp16, [2, 320, 1, 4096]> var_13925_cast = add(x = out_85_cast, y = var_13924_to_fp16)[name = tensor<string, []>("op_13925_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_13927_to_fp16 = const()[name = tensor<string, []>("op_13927_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214900928)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_307_cast = mul(x = var_13925_cast, y = var_13927_to_fp16)[name = tensor<string, []>("hidden_states_307_cast")];
            tensor<int32, [2]> var_13934 = const()[name = tensor<string, []>("op_13934"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13936 = const()[name = tensor<string, []>("op_13936"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214901632))), lut = tensor<fp16, [4]>([-0x1.95cp-3, -0x1.dc4p-5, 0x1.ea8p-5, 0x1.9b4p-3]), 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_13936, groups = var_12807, pad = q_57_pad_0, pad_type = q_57_pad_type_0, strides = var_13934, 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_13940 = const()[name = tensor<string, []>("op_13940"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13942 = const()[name = tensor<string, []>("op_13942"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214927296))), lut = tensor<fp16, [4]>([-0x1.908p-3, -0x1.b9p-5, 0x1.d7cp-5, 0x1.96cp-3]), 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_13942, groups = var_12807, pad = k_113_pad_0, pad_type = k_113_pad_type_0, strides = var_13940, 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_13946 = const()[name = tensor<string, []>("op_13946"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_13948 = const()[name = tensor<string, []>("op_13948"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214952960))), lut = tensor<fp16, [4]>([-0x1.828p-4, -0x1.a84p-6, 0x1.cc4p-6, 0x1.8bp-4]), 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_13948, groups = var_12807, pad = v_57_pad_0, pad_type = v_57_pad_type_0, strides = var_13946, 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_13952_begin_0 = const()[name = tensor<string, []>("op_13952_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13952_end_0 = const()[name = tensor<string, []>("op_13952_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13952_end_mask_0 = const()[name = tensor<string, []>("op_13952_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13952_cast = slice_by_index(begin = var_13952_begin_0, end = var_13952_end_0, end_mask = var_13952_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_13952_cast")];
            tensor<int32, [4]> var_13956_begin_0 = const()[name = tensor<string, []>("op_13956_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_13956_end_0 = const()[name = tensor<string, []>("op_13956_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_13956_end_mask_0 = const()[name = tensor<string, []>("op_13956_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13956_cast = slice_by_index(begin = var_13956_begin_0, end = var_13956_end_0, end_mask = var_13956_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_13956_cast")];
            tensor<int32, [4]> var_13960_begin_0 = const()[name = tensor<string, []>("op_13960_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_13960_end_0 = const()[name = tensor<string, []>("op_13960_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_13960_end_mask_0 = const()[name = tensor<string, []>("op_13960_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13960_cast = slice_by_index(begin = var_13960_begin_0, end = var_13960_end_0, end_mask = var_13960_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_13960_cast")];
            tensor<int32, [4]> var_13964_begin_0 = const()[name = tensor<string, []>("op_13964_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_13964_end_0 = const()[name = tensor<string, []>("op_13964_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_13964_end_mask_0 = const()[name = tensor<string, []>("op_13964_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13964_cast = slice_by_index(begin = var_13964_begin_0, end = var_13964_end_0, end_mask = var_13964_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_13964_cast")];
            tensor<int32, [4]> var_13968_begin_0 = const()[name = tensor<string, []>("op_13968_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_13968_end_0 = const()[name = tensor<string, []>("op_13968_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_13968_end_mask_0 = const()[name = tensor<string, []>("op_13968_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_13968_cast = slice_by_index(begin = var_13968_begin_0, end = var_13968_end_0, end_mask = var_13968_end_mask_0, x = q_57_cast)[name = tensor<string, []>("op_13968_cast")];
            tensor<int32, [4]> var_13971_begin_0 = const()[name = tensor<string, []>("op_13971_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13971_end_0 = const()[name = tensor<string, []>("op_13971_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13971_end_mask_0 = const()[name = tensor<string, []>("op_13971_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13971_cast = slice_by_index(begin = var_13971_begin_0, end = var_13971_end_0, end_mask = var_13971_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13971_cast")];
            tensor<int32, [4]> var_13972_begin_0 = const()[name = tensor<string, []>("op_13972_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13972_end_0 = const()[name = tensor<string, []>("op_13972_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13972_end_mask_0 = const()[name = tensor<string, []>("op_13972_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13972_cast = slice_by_index(begin = var_13972_begin_0, end = var_13972_end_0, end_mask = var_13972_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13972_cast")];
            tensor<int32, [4]> var_13973_begin_0 = const()[name = tensor<string, []>("op_13973_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13973_end_0 = const()[name = tensor<string, []>("op_13973_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13973_end_mask_0 = const()[name = tensor<string, []>("op_13973_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13973_cast = slice_by_index(begin = var_13973_begin_0, end = var_13973_end_0, end_mask = var_13973_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13973_cast")];
            tensor<int32, [4]> var_13974_begin_0 = const()[name = tensor<string, []>("op_13974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13974_end_0 = const()[name = tensor<string, []>("op_13974_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13974_end_mask_0 = const()[name = tensor<string, []>("op_13974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13974_cast = slice_by_index(begin = var_13974_begin_0, end = var_13974_end_0, end_mask = var_13974_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13974_cast")];
            tensor<int32, [4]> var_13975_begin_0 = const()[name = tensor<string, []>("op_13975_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13975_end_0 = const()[name = tensor<string, []>("op_13975_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13975_end_mask_0 = const()[name = tensor<string, []>("op_13975_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13975_cast = slice_by_index(begin = var_13975_begin_0, end = var_13975_end_0, end_mask = var_13975_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13975_cast")];
            tensor<int32, [4]> var_13976_begin_0 = const()[name = tensor<string, []>("op_13976_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13976_end_0 = const()[name = tensor<string, []>("op_13976_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13976_end_mask_0 = const()[name = tensor<string, []>("op_13976_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13976_cast = slice_by_index(begin = var_13976_begin_0, end = var_13976_end_0, end_mask = var_13976_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13976_cast")];
            tensor<int32, [4]> var_13977_begin_0 = const()[name = tensor<string, []>("op_13977_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13977_end_0 = const()[name = tensor<string, []>("op_13977_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13977_end_mask_0 = const()[name = tensor<string, []>("op_13977_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13977_cast = slice_by_index(begin = var_13977_begin_0, end = var_13977_end_0, end_mask = var_13977_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13977_cast")];
            tensor<int32, [4]> var_13978_begin_0 = const()[name = tensor<string, []>("op_13978_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13978_end_0 = const()[name = tensor<string, []>("op_13978_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13978_end_mask_0 = const()[name = tensor<string, []>("op_13978_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13978_cast = slice_by_index(begin = var_13978_begin_0, end = var_13978_end_0, end_mask = var_13978_end_mask_0, x = var_13952_cast)[name = tensor<string, []>("op_13978_cast")];
            tensor<int32, [4]> var_13979_begin_0 = const()[name = tensor<string, []>("op_13979_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13979_end_0 = const()[name = tensor<string, []>("op_13979_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13979_end_mask_0 = const()[name = tensor<string, []>("op_13979_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13979_cast = slice_by_index(begin = var_13979_begin_0, end = var_13979_end_0, end_mask = var_13979_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13979_cast")];
            tensor<int32, [4]> var_13980_begin_0 = const()[name = tensor<string, []>("op_13980_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13980_end_0 = const()[name = tensor<string, []>("op_13980_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13980_end_mask_0 = const()[name = tensor<string, []>("op_13980_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13980_cast = slice_by_index(begin = var_13980_begin_0, end = var_13980_end_0, end_mask = var_13980_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13980_cast")];
            tensor<int32, [4]> var_13981_begin_0 = const()[name = tensor<string, []>("op_13981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13981_end_0 = const()[name = tensor<string, []>("op_13981_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13981_end_mask_0 = const()[name = tensor<string, []>("op_13981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13981_cast = slice_by_index(begin = var_13981_begin_0, end = var_13981_end_0, end_mask = var_13981_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13981_cast")];
            tensor<int32, [4]> var_13982_begin_0 = const()[name = tensor<string, []>("op_13982_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13982_end_0 = const()[name = tensor<string, []>("op_13982_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13982_end_mask_0 = const()[name = tensor<string, []>("op_13982_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13982_cast = slice_by_index(begin = var_13982_begin_0, end = var_13982_end_0, end_mask = var_13982_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13982_cast")];
            tensor<int32, [4]> var_13983_begin_0 = const()[name = tensor<string, []>("op_13983_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13983_end_0 = const()[name = tensor<string, []>("op_13983_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13983_end_mask_0 = const()[name = tensor<string, []>("op_13983_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13983_cast = slice_by_index(begin = var_13983_begin_0, end = var_13983_end_0, end_mask = var_13983_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13983_cast")];
            tensor<int32, [4]> var_13984_begin_0 = const()[name = tensor<string, []>("op_13984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13984_end_0 = const()[name = tensor<string, []>("op_13984_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13984_end_mask_0 = const()[name = tensor<string, []>("op_13984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13984_cast = slice_by_index(begin = var_13984_begin_0, end = var_13984_end_0, end_mask = var_13984_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13984_cast")];
            tensor<int32, [4]> var_13985_begin_0 = const()[name = tensor<string, []>("op_13985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13985_end_0 = const()[name = tensor<string, []>("op_13985_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13985_end_mask_0 = const()[name = tensor<string, []>("op_13985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13985_cast = slice_by_index(begin = var_13985_begin_0, end = var_13985_end_0, end_mask = var_13985_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13985_cast")];
            tensor<int32, [4]> var_13986_begin_0 = const()[name = tensor<string, []>("op_13986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13986_end_0 = const()[name = tensor<string, []>("op_13986_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13986_end_mask_0 = const()[name = tensor<string, []>("op_13986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13986_cast = slice_by_index(begin = var_13986_begin_0, end = var_13986_end_0, end_mask = var_13986_end_mask_0, x = var_13956_cast)[name = tensor<string, []>("op_13986_cast")];
            tensor<int32, [4]> var_13987_begin_0 = const()[name = tensor<string, []>("op_13987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13987_end_0 = const()[name = tensor<string, []>("op_13987_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13987_end_mask_0 = const()[name = tensor<string, []>("op_13987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13987_cast = slice_by_index(begin = var_13987_begin_0, end = var_13987_end_0, end_mask = var_13987_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13987_cast")];
            tensor<int32, [4]> var_13988_begin_0 = const()[name = tensor<string, []>("op_13988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13988_end_0 = const()[name = tensor<string, []>("op_13988_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13988_end_mask_0 = const()[name = tensor<string, []>("op_13988_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13988_cast = slice_by_index(begin = var_13988_begin_0, end = var_13988_end_0, end_mask = var_13988_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13988_cast")];
            tensor<int32, [4]> var_13989_begin_0 = const()[name = tensor<string, []>("op_13989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13989_end_0 = const()[name = tensor<string, []>("op_13989_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13989_end_mask_0 = const()[name = tensor<string, []>("op_13989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13989_cast = slice_by_index(begin = var_13989_begin_0, end = var_13989_end_0, end_mask = var_13989_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13989_cast")];
            tensor<int32, [4]> var_13990_begin_0 = const()[name = tensor<string, []>("op_13990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13990_end_0 = const()[name = tensor<string, []>("op_13990_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13990_end_mask_0 = const()[name = tensor<string, []>("op_13990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13990_cast = slice_by_index(begin = var_13990_begin_0, end = var_13990_end_0, end_mask = var_13990_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13990_cast")];
            tensor<int32, [4]> var_13991_begin_0 = const()[name = tensor<string, []>("op_13991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13991_end_0 = const()[name = tensor<string, []>("op_13991_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13991_end_mask_0 = const()[name = tensor<string, []>("op_13991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13991_cast = slice_by_index(begin = var_13991_begin_0, end = var_13991_end_0, end_mask = var_13991_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13991_cast")];
            tensor<int32, [4]> var_13992_begin_0 = const()[name = tensor<string, []>("op_13992_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_13992_end_0 = const()[name = tensor<string, []>("op_13992_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_13992_end_mask_0 = const()[name = tensor<string, []>("op_13992_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13992_cast = slice_by_index(begin = var_13992_begin_0, end = var_13992_end_0, end_mask = var_13992_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13992_cast")];
            tensor<int32, [4]> var_13993_begin_0 = const()[name = tensor<string, []>("op_13993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_13993_end_0 = const()[name = tensor<string, []>("op_13993_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_13993_end_mask_0 = const()[name = tensor<string, []>("op_13993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13993_cast = slice_by_index(begin = var_13993_begin_0, end = var_13993_end_0, end_mask = var_13993_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13993_cast")];
            tensor<int32, [4]> var_13994_begin_0 = const()[name = tensor<string, []>("op_13994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_13994_end_0 = const()[name = tensor<string, []>("op_13994_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_13994_end_mask_0 = const()[name = tensor<string, []>("op_13994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13994_cast = slice_by_index(begin = var_13994_begin_0, end = var_13994_end_0, end_mask = var_13994_end_mask_0, x = var_13960_cast)[name = tensor<string, []>("op_13994_cast")];
            tensor<int32, [4]> var_13995_begin_0 = const()[name = tensor<string, []>("op_13995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_13995_end_0 = const()[name = tensor<string, []>("op_13995_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_13995_end_mask_0 = const()[name = tensor<string, []>("op_13995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13995_cast = slice_by_index(begin = var_13995_begin_0, end = var_13995_end_0, end_mask = var_13995_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_13995_cast")];
            tensor<int32, [4]> var_13996_begin_0 = const()[name = tensor<string, []>("op_13996_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_13996_end_0 = const()[name = tensor<string, []>("op_13996_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_13996_end_mask_0 = const()[name = tensor<string, []>("op_13996_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13996_cast = slice_by_index(begin = var_13996_begin_0, end = var_13996_end_0, end_mask = var_13996_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_13996_cast")];
            tensor<int32, [4]> var_13997_begin_0 = const()[name = tensor<string, []>("op_13997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_13997_end_0 = const()[name = tensor<string, []>("op_13997_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_13997_end_mask_0 = const()[name = tensor<string, []>("op_13997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13997_cast = slice_by_index(begin = var_13997_begin_0, end = var_13997_end_0, end_mask = var_13997_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_13997_cast")];
            tensor<int32, [4]> var_13998_begin_0 = const()[name = tensor<string, []>("op_13998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_13998_end_0 = const()[name = tensor<string, []>("op_13998_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_13998_end_mask_0 = const()[name = tensor<string, []>("op_13998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13998_cast = slice_by_index(begin = var_13998_begin_0, end = var_13998_end_0, end_mask = var_13998_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_13998_cast")];
            tensor<int32, [4]> var_13999_begin_0 = const()[name = tensor<string, []>("op_13999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_13999_end_0 = const()[name = tensor<string, []>("op_13999_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_13999_end_mask_0 = const()[name = tensor<string, []>("op_13999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_13999_cast = slice_by_index(begin = var_13999_begin_0, end = var_13999_end_0, end_mask = var_13999_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_13999_cast")];
            tensor<int32, [4]> var_14000_begin_0 = const()[name = tensor<string, []>("op_14000_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14000_end_0 = const()[name = tensor<string, []>("op_14000_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14000_end_mask_0 = const()[name = tensor<string, []>("op_14000_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14000_cast = slice_by_index(begin = var_14000_begin_0, end = var_14000_end_0, end_mask = var_14000_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_14000_cast")];
            tensor<int32, [4]> var_14001_begin_0 = const()[name = tensor<string, []>("op_14001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14001_end_0 = const()[name = tensor<string, []>("op_14001_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14001_end_mask_0 = const()[name = tensor<string, []>("op_14001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14001_cast = slice_by_index(begin = var_14001_begin_0, end = var_14001_end_0, end_mask = var_14001_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_14001_cast")];
            tensor<int32, [4]> var_14002_begin_0 = const()[name = tensor<string, []>("op_14002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14002_end_0 = const()[name = tensor<string, []>("op_14002_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14002_end_mask_0 = const()[name = tensor<string, []>("op_14002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14002_cast = slice_by_index(begin = var_14002_begin_0, end = var_14002_end_0, end_mask = var_14002_end_mask_0, x = var_13964_cast)[name = tensor<string, []>("op_14002_cast")];
            tensor<int32, [4]> var_14003_begin_0 = const()[name = tensor<string, []>("op_14003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14003_end_0 = const()[name = tensor<string, []>("op_14003_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14003_end_mask_0 = const()[name = tensor<string, []>("op_14003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14003_cast = slice_by_index(begin = var_14003_begin_0, end = var_14003_end_0, end_mask = var_14003_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14003_cast")];
            tensor<int32, [4]> var_14004_begin_0 = const()[name = tensor<string, []>("op_14004_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14004_end_0 = const()[name = tensor<string, []>("op_14004_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14004_end_mask_0 = const()[name = tensor<string, []>("op_14004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14004_cast = slice_by_index(begin = var_14004_begin_0, end = var_14004_end_0, end_mask = var_14004_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14004_cast")];
            tensor<int32, [4]> var_14005_begin_0 = const()[name = tensor<string, []>("op_14005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14005_end_0 = const()[name = tensor<string, []>("op_14005_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14005_end_mask_0 = const()[name = tensor<string, []>("op_14005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14005_cast = slice_by_index(begin = var_14005_begin_0, end = var_14005_end_0, end_mask = var_14005_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14005_cast")];
            tensor<int32, [4]> var_14006_begin_0 = const()[name = tensor<string, []>("op_14006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14006_end_0 = const()[name = tensor<string, []>("op_14006_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14006_end_mask_0 = const()[name = tensor<string, []>("op_14006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14006_cast = slice_by_index(begin = var_14006_begin_0, end = var_14006_end_0, end_mask = var_14006_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14006_cast")];
            tensor<int32, [4]> var_14007_begin_0 = const()[name = tensor<string, []>("op_14007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14007_end_0 = const()[name = tensor<string, []>("op_14007_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14007_end_mask_0 = const()[name = tensor<string, []>("op_14007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14007_cast = slice_by_index(begin = var_14007_begin_0, end = var_14007_end_0, end_mask = var_14007_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14007_cast")];
            tensor<int32, [4]> var_14008_begin_0 = const()[name = tensor<string, []>("op_14008_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14008_end_0 = const()[name = tensor<string, []>("op_14008_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14008_end_mask_0 = const()[name = tensor<string, []>("op_14008_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14008_cast = slice_by_index(begin = var_14008_begin_0, end = var_14008_end_0, end_mask = var_14008_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14008_cast")];
            tensor<int32, [4]> var_14009_begin_0 = const()[name = tensor<string, []>("op_14009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14009_end_0 = const()[name = tensor<string, []>("op_14009_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14009_end_mask_0 = const()[name = tensor<string, []>("op_14009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14009_cast = slice_by_index(begin = var_14009_begin_0, end = var_14009_end_0, end_mask = var_14009_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14009_cast")];
            tensor<int32, [4]> var_14010_begin_0 = const()[name = tensor<string, []>("op_14010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14010_end_0 = const()[name = tensor<string, []>("op_14010_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14010_end_mask_0 = const()[name = tensor<string, []>("op_14010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14010_cast = slice_by_index(begin = var_14010_begin_0, end = var_14010_end_0, end_mask = var_14010_end_mask_0, x = var_13968_cast)[name = tensor<string, []>("op_14010_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_14015_begin_0 = const()[name = tensor<string, []>("op_14015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14015_end_0 = const()[name = tensor<string, []>("op_14015_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_14015_end_mask_0 = const()[name = tensor<string, []>("op_14015_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, 64]> var_14015_cast = slice_by_index(begin = var_14015_begin_0, end = var_14015_end_0, end_mask = var_14015_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_14015_cast")];
            tensor<int32, [4]> var_14019_begin_0 = const()[name = tensor<string, []>("op_14019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_14019_end_0 = const()[name = tensor<string, []>("op_14019_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_14019_end_mask_0 = const()[name = tensor<string, []>("op_14019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_14019_cast = slice_by_index(begin = var_14019_begin_0, end = var_14019_end_0, end_mask = var_14019_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_14019_cast")];
            tensor<int32, [4]> var_14023_begin_0 = const()[name = tensor<string, []>("op_14023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_14023_end_0 = const()[name = tensor<string, []>("op_14023_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_14023_end_mask_0 = const()[name = tensor<string, []>("op_14023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_14023_cast = slice_by_index(begin = var_14023_begin_0, end = var_14023_end_0, end_mask = var_14023_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_14023_cast")];
            tensor<int32, [4]> var_14027_begin_0 = const()[name = tensor<string, []>("op_14027_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_14027_end_0 = const()[name = tensor<string, []>("op_14027_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_14027_end_mask_0 = const()[name = tensor<string, []>("op_14027_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_14027_cast = slice_by_index(begin = var_14027_begin_0, end = var_14027_end_0, end_mask = var_14027_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_14027_cast")];
            tensor<int32, [4]> var_14031_begin_0 = const()[name = tensor<string, []>("op_14031_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_14031_end_0 = const()[name = tensor<string, []>("op_14031_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_14031_end_mask_0 = const()[name = tensor<string, []>("op_14031_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_14031_cast = slice_by_index(begin = var_14031_begin_0, end = var_14031_end_0, end_mask = var_14031_end_mask_0, x = transpose_3)[name = tensor<string, []>("op_14031_cast")];
            tensor<int32, [4]> var_14033_begin_0 = const()[name = tensor<string, []>("op_14033_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14033_end_0 = const()[name = tensor<string, []>("op_14033_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14033_end_mask_0 = const()[name = tensor<string, []>("op_14033_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14033_cast = slice_by_index(begin = var_14033_begin_0, end = var_14033_end_0, end_mask = var_14033_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_14033_cast")];
            tensor<int32, [4]> var_14037_begin_0 = const()[name = tensor<string, []>("op_14037_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14037_end_0 = const()[name = tensor<string, []>("op_14037_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_14037_end_mask_0 = const()[name = tensor<string, []>("op_14037_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14037_cast = slice_by_index(begin = var_14037_begin_0, end = var_14037_end_0, end_mask = var_14037_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_14037_cast")];
            tensor<int32, [4]> var_14041_begin_0 = const()[name = tensor<string, []>("op_14041_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14041_end_0 = const()[name = tensor<string, []>("op_14041_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_14041_end_mask_0 = const()[name = tensor<string, []>("op_14041_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14041_cast = slice_by_index(begin = var_14041_begin_0, end = var_14041_end_0, end_mask = var_14041_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_14041_cast")];
            tensor<int32, [4]> var_14045_begin_0 = const()[name = tensor<string, []>("op_14045_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14045_end_0 = const()[name = tensor<string, []>("op_14045_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_14045_end_mask_0 = const()[name = tensor<string, []>("op_14045_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14045_cast = slice_by_index(begin = var_14045_begin_0, end = var_14045_end_0, end_mask = var_14045_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_14045_cast")];
            tensor<int32, [4]> var_14049_begin_0 = const()[name = tensor<string, []>("op_14049_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14049_end_0 = const()[name = tensor<string, []>("op_14049_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_14049_end_mask_0 = const()[name = tensor<string, []>("op_14049_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14049_cast = slice_by_index(begin = var_14049_begin_0, end = var_14049_end_0, end_mask = var_14049_end_mask_0, x = v_57_cast)[name = tensor<string, []>("op_14049_cast")];
            tensor<string, []> var_14053_equation_0 = const()[name = tensor<string, []>("op_14053_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14053_cast = einsum(equation = var_14053_equation_0, values = (var_14015_cast, var_13971_cast))[name = tensor<string, []>("op_14053_cast")];
            tensor<fp16, []> var_14054_to_fp16 = const()[name = tensor<string, []>("op_14054_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_881_cast = mul(x = var_14053_cast, y = var_14054_to_fp16)[name = tensor<string, []>("aw_chunk_881_cast")];
            tensor<string, []> var_14057_equation_0 = const()[name = tensor<string, []>("op_14057_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14057_cast = einsum(equation = var_14057_equation_0, values = (var_14015_cast, var_13972_cast))[name = tensor<string, []>("op_14057_cast")];
            tensor<fp16, []> var_14058_to_fp16 = const()[name = tensor<string, []>("op_14058_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_883_cast = mul(x = var_14057_cast, y = var_14058_to_fp16)[name = tensor<string, []>("aw_chunk_883_cast")];
            tensor<string, []> var_14061_equation_0 = const()[name = tensor<string, []>("op_14061_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14061_cast = einsum(equation = var_14061_equation_0, values = (var_14015_cast, var_13973_cast))[name = tensor<string, []>("op_14061_cast")];
            tensor<fp16, []> var_14062_to_fp16 = const()[name = tensor<string, []>("op_14062_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_885_cast = mul(x = var_14061_cast, y = var_14062_to_fp16)[name = tensor<string, []>("aw_chunk_885_cast")];
            tensor<string, []> var_14065_equation_0 = const()[name = tensor<string, []>("op_14065_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14065_cast = einsum(equation = var_14065_equation_0, values = (var_14015_cast, var_13974_cast))[name = tensor<string, []>("op_14065_cast")];
            tensor<fp16, []> var_14066_to_fp16 = const()[name = tensor<string, []>("op_14066_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_887_cast = mul(x = var_14065_cast, y = var_14066_to_fp16)[name = tensor<string, []>("aw_chunk_887_cast")];
            tensor<string, []> var_14069_equation_0 = const()[name = tensor<string, []>("op_14069_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14069_cast = einsum(equation = var_14069_equation_0, values = (var_14015_cast, var_13975_cast))[name = tensor<string, []>("op_14069_cast")];
            tensor<fp16, []> var_14070_to_fp16 = const()[name = tensor<string, []>("op_14070_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_889_cast = mul(x = var_14069_cast, y = var_14070_to_fp16)[name = tensor<string, []>("aw_chunk_889_cast")];
            tensor<string, []> var_14073_equation_0 = const()[name = tensor<string, []>("op_14073_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14073_cast = einsum(equation = var_14073_equation_0, values = (var_14015_cast, var_13976_cast))[name = tensor<string, []>("op_14073_cast")];
            tensor<fp16, []> var_14074_to_fp16 = const()[name = tensor<string, []>("op_14074_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_891_cast = mul(x = var_14073_cast, y = var_14074_to_fp16)[name = tensor<string, []>("aw_chunk_891_cast")];
            tensor<string, []> var_14077_equation_0 = const()[name = tensor<string, []>("op_14077_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14077_cast = einsum(equation = var_14077_equation_0, values = (var_14015_cast, var_13977_cast))[name = tensor<string, []>("op_14077_cast")];
            tensor<fp16, []> var_14078_to_fp16 = const()[name = tensor<string, []>("op_14078_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_893_cast = mul(x = var_14077_cast, y = var_14078_to_fp16)[name = tensor<string, []>("aw_chunk_893_cast")];
            tensor<string, []> var_14081_equation_0 = const()[name = tensor<string, []>("op_14081_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14081_cast = einsum(equation = var_14081_equation_0, values = (var_14015_cast, var_13978_cast))[name = tensor<string, []>("op_14081_cast")];
            tensor<fp16, []> var_14082_to_fp16 = const()[name = tensor<string, []>("op_14082_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_895_cast = mul(x = var_14081_cast, y = var_14082_to_fp16)[name = tensor<string, []>("aw_chunk_895_cast")];
            tensor<string, []> var_14085_equation_0 = const()[name = tensor<string, []>("op_14085_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14085_cast = einsum(equation = var_14085_equation_0, values = (var_14019_cast, var_13979_cast))[name = tensor<string, []>("op_14085_cast")];
            tensor<fp16, []> var_14086_to_fp16 = const()[name = tensor<string, []>("op_14086_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_897_cast = mul(x = var_14085_cast, y = var_14086_to_fp16)[name = tensor<string, []>("aw_chunk_897_cast")];
            tensor<string, []> var_14089_equation_0 = const()[name = tensor<string, []>("op_14089_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14089_cast = einsum(equation = var_14089_equation_0, values = (var_14019_cast, var_13980_cast))[name = tensor<string, []>("op_14089_cast")];
            tensor<fp16, []> var_14090_to_fp16 = const()[name = tensor<string, []>("op_14090_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_899_cast = mul(x = var_14089_cast, y = var_14090_to_fp16)[name = tensor<string, []>("aw_chunk_899_cast")];
            tensor<string, []> var_14093_equation_0 = const()[name = tensor<string, []>("op_14093_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14093_cast = einsum(equation = var_14093_equation_0, values = (var_14019_cast, var_13981_cast))[name = tensor<string, []>("op_14093_cast")];
            tensor<fp16, []> var_14094_to_fp16 = const()[name = tensor<string, []>("op_14094_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_901_cast = mul(x = var_14093_cast, y = var_14094_to_fp16)[name = tensor<string, []>("aw_chunk_901_cast")];
            tensor<string, []> var_14097_equation_0 = const()[name = tensor<string, []>("op_14097_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14097_cast = einsum(equation = var_14097_equation_0, values = (var_14019_cast, var_13982_cast))[name = tensor<string, []>("op_14097_cast")];
            tensor<fp16, []> var_14098_to_fp16 = const()[name = tensor<string, []>("op_14098_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_903_cast = mul(x = var_14097_cast, y = var_14098_to_fp16)[name = tensor<string, []>("aw_chunk_903_cast")];
            tensor<string, []> var_14101_equation_0 = const()[name = tensor<string, []>("op_14101_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14101_cast = einsum(equation = var_14101_equation_0, values = (var_14019_cast, var_13983_cast))[name = tensor<string, []>("op_14101_cast")];
            tensor<fp16, []> var_14102_to_fp16 = const()[name = tensor<string, []>("op_14102_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_905_cast = mul(x = var_14101_cast, y = var_14102_to_fp16)[name = tensor<string, []>("aw_chunk_905_cast")];
            tensor<string, []> var_14105_equation_0 = const()[name = tensor<string, []>("op_14105_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14105_cast = einsum(equation = var_14105_equation_0, values = (var_14019_cast, var_13984_cast))[name = tensor<string, []>("op_14105_cast")];
            tensor<fp16, []> var_14106_to_fp16 = const()[name = tensor<string, []>("op_14106_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_907_cast = mul(x = var_14105_cast, y = var_14106_to_fp16)[name = tensor<string, []>("aw_chunk_907_cast")];
            tensor<string, []> var_14109_equation_0 = const()[name = tensor<string, []>("op_14109_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14109_cast = einsum(equation = var_14109_equation_0, values = (var_14019_cast, var_13985_cast))[name = tensor<string, []>("op_14109_cast")];
            tensor<fp16, []> var_14110_to_fp16 = const()[name = tensor<string, []>("op_14110_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_909_cast = mul(x = var_14109_cast, y = var_14110_to_fp16)[name = tensor<string, []>("aw_chunk_909_cast")];
            tensor<string, []> var_14113_equation_0 = const()[name = tensor<string, []>("op_14113_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14113_cast = einsum(equation = var_14113_equation_0, values = (var_14019_cast, var_13986_cast))[name = tensor<string, []>("op_14113_cast")];
            tensor<fp16, []> var_14114_to_fp16 = const()[name = tensor<string, []>("op_14114_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_911_cast = mul(x = var_14113_cast, y = var_14114_to_fp16)[name = tensor<string, []>("aw_chunk_911_cast")];
            tensor<string, []> var_14117_equation_0 = const()[name = tensor<string, []>("op_14117_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14117_cast = einsum(equation = var_14117_equation_0, values = (var_14023_cast, var_13987_cast))[name = tensor<string, []>("op_14117_cast")];
            tensor<fp16, []> var_14118_to_fp16 = const()[name = tensor<string, []>("op_14118_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_913_cast = mul(x = var_14117_cast, y = var_14118_to_fp16)[name = tensor<string, []>("aw_chunk_913_cast")];
            tensor<string, []> var_14121_equation_0 = const()[name = tensor<string, []>("op_14121_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14121_cast = einsum(equation = var_14121_equation_0, values = (var_14023_cast, var_13988_cast))[name = tensor<string, []>("op_14121_cast")];
            tensor<fp16, []> var_14122_to_fp16 = const()[name = tensor<string, []>("op_14122_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_915_cast = mul(x = var_14121_cast, y = var_14122_to_fp16)[name = tensor<string, []>("aw_chunk_915_cast")];
            tensor<string, []> var_14125_equation_0 = const()[name = tensor<string, []>("op_14125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14125_cast = einsum(equation = var_14125_equation_0, values = (var_14023_cast, var_13989_cast))[name = tensor<string, []>("op_14125_cast")];
            tensor<fp16, []> var_14126_to_fp16 = const()[name = tensor<string, []>("op_14126_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_917_cast = mul(x = var_14125_cast, y = var_14126_to_fp16)[name = tensor<string, []>("aw_chunk_917_cast")];
            tensor<string, []> var_14129_equation_0 = const()[name = tensor<string, []>("op_14129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14129_cast = einsum(equation = var_14129_equation_0, values = (var_14023_cast, var_13990_cast))[name = tensor<string, []>("op_14129_cast")];
            tensor<fp16, []> var_14130_to_fp16 = const()[name = tensor<string, []>("op_14130_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_919_cast = mul(x = var_14129_cast, y = var_14130_to_fp16)[name = tensor<string, []>("aw_chunk_919_cast")];
            tensor<string, []> var_14133_equation_0 = const()[name = tensor<string, []>("op_14133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14133_cast = einsum(equation = var_14133_equation_0, values = (var_14023_cast, var_13991_cast))[name = tensor<string, []>("op_14133_cast")];
            tensor<fp16, []> var_14134_to_fp16 = const()[name = tensor<string, []>("op_14134_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_921_cast = mul(x = var_14133_cast, y = var_14134_to_fp16)[name = tensor<string, []>("aw_chunk_921_cast")];
            tensor<string, []> var_14137_equation_0 = const()[name = tensor<string, []>("op_14137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14137_cast = einsum(equation = var_14137_equation_0, values = (var_14023_cast, var_13992_cast))[name = tensor<string, []>("op_14137_cast")];
            tensor<fp16, []> var_14138_to_fp16 = const()[name = tensor<string, []>("op_14138_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_923_cast = mul(x = var_14137_cast, y = var_14138_to_fp16)[name = tensor<string, []>("aw_chunk_923_cast")];
            tensor<string, []> var_14141_equation_0 = const()[name = tensor<string, []>("op_14141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14141_cast = einsum(equation = var_14141_equation_0, values = (var_14023_cast, var_13993_cast))[name = tensor<string, []>("op_14141_cast")];
            tensor<fp16, []> var_14142_to_fp16 = const()[name = tensor<string, []>("op_14142_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_925_cast = mul(x = var_14141_cast, y = var_14142_to_fp16)[name = tensor<string, []>("aw_chunk_925_cast")];
            tensor<string, []> var_14145_equation_0 = const()[name = tensor<string, []>("op_14145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14145_cast = einsum(equation = var_14145_equation_0, values = (var_14023_cast, var_13994_cast))[name = tensor<string, []>("op_14145_cast")];
            tensor<fp16, []> var_14146_to_fp16 = const()[name = tensor<string, []>("op_14146_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_927_cast = mul(x = var_14145_cast, y = var_14146_to_fp16)[name = tensor<string, []>("aw_chunk_927_cast")];
            tensor<string, []> var_14149_equation_0 = const()[name = tensor<string, []>("op_14149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14149_cast = einsum(equation = var_14149_equation_0, values = (var_14027_cast, var_13995_cast))[name = tensor<string, []>("op_14149_cast")];
            tensor<fp16, []> var_14150_to_fp16 = const()[name = tensor<string, []>("op_14150_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_929_cast = mul(x = var_14149_cast, y = var_14150_to_fp16)[name = tensor<string, []>("aw_chunk_929_cast")];
            tensor<string, []> var_14153_equation_0 = const()[name = tensor<string, []>("op_14153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14153_cast = einsum(equation = var_14153_equation_0, values = (var_14027_cast, var_13996_cast))[name = tensor<string, []>("op_14153_cast")];
            tensor<fp16, []> var_14154_to_fp16 = const()[name = tensor<string, []>("op_14154_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_931_cast = mul(x = var_14153_cast, y = var_14154_to_fp16)[name = tensor<string, []>("aw_chunk_931_cast")];
            tensor<string, []> var_14157_equation_0 = const()[name = tensor<string, []>("op_14157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14157_cast = einsum(equation = var_14157_equation_0, values = (var_14027_cast, var_13997_cast))[name = tensor<string, []>("op_14157_cast")];
            tensor<fp16, []> var_14158_to_fp16 = const()[name = tensor<string, []>("op_14158_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_933_cast = mul(x = var_14157_cast, y = var_14158_to_fp16)[name = tensor<string, []>("aw_chunk_933_cast")];
            tensor<string, []> var_14161_equation_0 = const()[name = tensor<string, []>("op_14161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14161_cast = einsum(equation = var_14161_equation_0, values = (var_14027_cast, var_13998_cast))[name = tensor<string, []>("op_14161_cast")];
            tensor<fp16, []> var_14162_to_fp16 = const()[name = tensor<string, []>("op_14162_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_935_cast = mul(x = var_14161_cast, y = var_14162_to_fp16)[name = tensor<string, []>("aw_chunk_935_cast")];
            tensor<string, []> var_14165_equation_0 = const()[name = tensor<string, []>("op_14165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14165_cast = einsum(equation = var_14165_equation_0, values = (var_14027_cast, var_13999_cast))[name = tensor<string, []>("op_14165_cast")];
            tensor<fp16, []> var_14166_to_fp16 = const()[name = tensor<string, []>("op_14166_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_937_cast = mul(x = var_14165_cast, y = var_14166_to_fp16)[name = tensor<string, []>("aw_chunk_937_cast")];
            tensor<string, []> var_14169_equation_0 = const()[name = tensor<string, []>("op_14169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14169_cast = einsum(equation = var_14169_equation_0, values = (var_14027_cast, var_14000_cast))[name = tensor<string, []>("op_14169_cast")];
            tensor<fp16, []> var_14170_to_fp16 = const()[name = tensor<string, []>("op_14170_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_939_cast = mul(x = var_14169_cast, y = var_14170_to_fp16)[name = tensor<string, []>("aw_chunk_939_cast")];
            tensor<string, []> var_14173_equation_0 = const()[name = tensor<string, []>("op_14173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14173_cast = einsum(equation = var_14173_equation_0, values = (var_14027_cast, var_14001_cast))[name = tensor<string, []>("op_14173_cast")];
            tensor<fp16, []> var_14174_to_fp16 = const()[name = tensor<string, []>("op_14174_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_941_cast = mul(x = var_14173_cast, y = var_14174_to_fp16)[name = tensor<string, []>("aw_chunk_941_cast")];
            tensor<string, []> var_14177_equation_0 = const()[name = tensor<string, []>("op_14177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14177_cast = einsum(equation = var_14177_equation_0, values = (var_14027_cast, var_14002_cast))[name = tensor<string, []>("op_14177_cast")];
            tensor<fp16, []> var_14178_to_fp16 = const()[name = tensor<string, []>("op_14178_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_943_cast = mul(x = var_14177_cast, y = var_14178_to_fp16)[name = tensor<string, []>("aw_chunk_943_cast")];
            tensor<string, []> var_14181_equation_0 = const()[name = tensor<string, []>("op_14181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14181_cast = einsum(equation = var_14181_equation_0, values = (var_14031_cast, var_14003_cast))[name = tensor<string, []>("op_14181_cast")];
            tensor<fp16, []> var_14182_to_fp16 = const()[name = tensor<string, []>("op_14182_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_945_cast = mul(x = var_14181_cast, y = var_14182_to_fp16)[name = tensor<string, []>("aw_chunk_945_cast")];
            tensor<string, []> var_14185_equation_0 = const()[name = tensor<string, []>("op_14185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14185_cast = einsum(equation = var_14185_equation_0, values = (var_14031_cast, var_14004_cast))[name = tensor<string, []>("op_14185_cast")];
            tensor<fp16, []> var_14186_to_fp16 = const()[name = tensor<string, []>("op_14186_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_947_cast = mul(x = var_14185_cast, y = var_14186_to_fp16)[name = tensor<string, []>("aw_chunk_947_cast")];
            tensor<string, []> var_14189_equation_0 = const()[name = tensor<string, []>("op_14189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14189_cast = einsum(equation = var_14189_equation_0, values = (var_14031_cast, var_14005_cast))[name = tensor<string, []>("op_14189_cast")];
            tensor<fp16, []> var_14190_to_fp16 = const()[name = tensor<string, []>("op_14190_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_949_cast = mul(x = var_14189_cast, y = var_14190_to_fp16)[name = tensor<string, []>("aw_chunk_949_cast")];
            tensor<string, []> var_14193_equation_0 = const()[name = tensor<string, []>("op_14193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14193_cast = einsum(equation = var_14193_equation_0, values = (var_14031_cast, var_14006_cast))[name = tensor<string, []>("op_14193_cast")];
            tensor<fp16, []> var_14194_to_fp16 = const()[name = tensor<string, []>("op_14194_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_951_cast = mul(x = var_14193_cast, y = var_14194_to_fp16)[name = tensor<string, []>("aw_chunk_951_cast")];
            tensor<string, []> var_14197_equation_0 = const()[name = tensor<string, []>("op_14197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14197_cast = einsum(equation = var_14197_equation_0, values = (var_14031_cast, var_14007_cast))[name = tensor<string, []>("op_14197_cast")];
            tensor<fp16, []> var_14198_to_fp16 = const()[name = tensor<string, []>("op_14198_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_953_cast = mul(x = var_14197_cast, y = var_14198_to_fp16)[name = tensor<string, []>("aw_chunk_953_cast")];
            tensor<string, []> var_14201_equation_0 = const()[name = tensor<string, []>("op_14201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14201_cast = einsum(equation = var_14201_equation_0, values = (var_14031_cast, var_14008_cast))[name = tensor<string, []>("op_14201_cast")];
            tensor<fp16, []> var_14202_to_fp16 = const()[name = tensor<string, []>("op_14202_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_955_cast = mul(x = var_14201_cast, y = var_14202_to_fp16)[name = tensor<string, []>("aw_chunk_955_cast")];
            tensor<string, []> var_14205_equation_0 = const()[name = tensor<string, []>("op_14205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14205_cast = einsum(equation = var_14205_equation_0, values = (var_14031_cast, var_14009_cast))[name = tensor<string, []>("op_14205_cast")];
            tensor<fp16, []> var_14206_to_fp16 = const()[name = tensor<string, []>("op_14206_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_957_cast = mul(x = var_14205_cast, y = var_14206_to_fp16)[name = tensor<string, []>("aw_chunk_957_cast")];
            tensor<string, []> var_14209_equation_0 = const()[name = tensor<string, []>("op_14209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_14209_cast = einsum(equation = var_14209_equation_0, values = (var_14031_cast, var_14010_cast))[name = tensor<string, []>("op_14209_cast")];
            tensor<fp16, []> var_14210_to_fp16 = const()[name = tensor<string, []>("op_14210_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_959_cast = mul(x = var_14209_cast, y = var_14210_to_fp16)[name = tensor<string, []>("aw_chunk_959_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14212_cast = softmax(axis = var_12807, x = aw_chunk_881_cast)[name = tensor<string, []>("op_14212_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14213_cast = softmax(axis = var_12807, x = aw_chunk_883_cast)[name = tensor<string, []>("op_14213_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14214_cast = softmax(axis = var_12807, x = aw_chunk_885_cast)[name = tensor<string, []>("op_14214_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14215_cast = softmax(axis = var_12807, x = aw_chunk_887_cast)[name = tensor<string, []>("op_14215_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14216_cast = softmax(axis = var_12807, x = aw_chunk_889_cast)[name = tensor<string, []>("op_14216_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14217_cast = softmax(axis = var_12807, x = aw_chunk_891_cast)[name = tensor<string, []>("op_14217_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14218_cast = softmax(axis = var_12807, x = aw_chunk_893_cast)[name = tensor<string, []>("op_14218_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14219_cast = softmax(axis = var_12807, x = aw_chunk_895_cast)[name = tensor<string, []>("op_14219_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14220_cast = softmax(axis = var_12807, x = aw_chunk_897_cast)[name = tensor<string, []>("op_14220_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14221_cast = softmax(axis = var_12807, x = aw_chunk_899_cast)[name = tensor<string, []>("op_14221_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14222_cast = softmax(axis = var_12807, x = aw_chunk_901_cast)[name = tensor<string, []>("op_14222_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14223_cast = softmax(axis = var_12807, x = aw_chunk_903_cast)[name = tensor<string, []>("op_14223_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14224_cast = softmax(axis = var_12807, x = aw_chunk_905_cast)[name = tensor<string, []>("op_14224_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14225_cast = softmax(axis = var_12807, x = aw_chunk_907_cast)[name = tensor<string, []>("op_14225_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14226_cast = softmax(axis = var_12807, x = aw_chunk_909_cast)[name = tensor<string, []>("op_14226_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14227_cast = softmax(axis = var_12807, x = aw_chunk_911_cast)[name = tensor<string, []>("op_14227_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14228_cast = softmax(axis = var_12807, x = aw_chunk_913_cast)[name = tensor<string, []>("op_14228_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14229_cast = softmax(axis = var_12807, x = aw_chunk_915_cast)[name = tensor<string, []>("op_14229_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14230_cast = softmax(axis = var_12807, x = aw_chunk_917_cast)[name = tensor<string, []>("op_14230_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14231_cast = softmax(axis = var_12807, x = aw_chunk_919_cast)[name = tensor<string, []>("op_14231_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14232_cast = softmax(axis = var_12807, x = aw_chunk_921_cast)[name = tensor<string, []>("op_14232_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14233_cast = softmax(axis = var_12807, x = aw_chunk_923_cast)[name = tensor<string, []>("op_14233_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14234_cast = softmax(axis = var_12807, x = aw_chunk_925_cast)[name = tensor<string, []>("op_14234_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14235_cast = softmax(axis = var_12807, x = aw_chunk_927_cast)[name = tensor<string, []>("op_14235_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14236_cast = softmax(axis = var_12807, x = aw_chunk_929_cast)[name = tensor<string, []>("op_14236_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14237_cast = softmax(axis = var_12807, x = aw_chunk_931_cast)[name = tensor<string, []>("op_14237_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14238_cast = softmax(axis = var_12807, x = aw_chunk_933_cast)[name = tensor<string, []>("op_14238_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14239_cast = softmax(axis = var_12807, x = aw_chunk_935_cast)[name = tensor<string, []>("op_14239_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14240_cast = softmax(axis = var_12807, x = aw_chunk_937_cast)[name = tensor<string, []>("op_14240_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14241_cast = softmax(axis = var_12807, x = aw_chunk_939_cast)[name = tensor<string, []>("op_14241_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14242_cast = softmax(axis = var_12807, x = aw_chunk_941_cast)[name = tensor<string, []>("op_14242_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14243_cast = softmax(axis = var_12807, x = aw_chunk_943_cast)[name = tensor<string, []>("op_14243_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14244_cast = softmax(axis = var_12807, x = aw_chunk_945_cast)[name = tensor<string, []>("op_14244_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14245_cast = softmax(axis = var_12807, x = aw_chunk_947_cast)[name = tensor<string, []>("op_14245_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14246_cast = softmax(axis = var_12807, x = aw_chunk_949_cast)[name = tensor<string, []>("op_14246_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14247_cast = softmax(axis = var_12807, x = aw_chunk_951_cast)[name = tensor<string, []>("op_14247_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14248_cast = softmax(axis = var_12807, x = aw_chunk_953_cast)[name = tensor<string, []>("op_14248_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14249_cast = softmax(axis = var_12807, x = aw_chunk_955_cast)[name = tensor<string, []>("op_14249_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14250_cast = softmax(axis = var_12807, x = aw_chunk_957_cast)[name = tensor<string, []>("op_14250_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_14251_cast = softmax(axis = var_12807, x = aw_chunk_959_cast)[name = tensor<string, []>("op_14251_cast")];
            tensor<string, []> var_14253_equation_0 = const()[name = tensor<string, []>("op_14253_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14253_cast = einsum(equation = var_14253_equation_0, values = (var_14033_cast, var_14212_cast))[name = tensor<string, []>("op_14253_cast")];
            tensor<string, []> var_14255_equation_0 = const()[name = tensor<string, []>("op_14255_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14255_cast = einsum(equation = var_14255_equation_0, values = (var_14033_cast, var_14213_cast))[name = tensor<string, []>("op_14255_cast")];
            tensor<string, []> var_14257_equation_0 = const()[name = tensor<string, []>("op_14257_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14257_cast = einsum(equation = var_14257_equation_0, values = (var_14033_cast, var_14214_cast))[name = tensor<string, []>("op_14257_cast")];
            tensor<string, []> var_14259_equation_0 = const()[name = tensor<string, []>("op_14259_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14259_cast = einsum(equation = var_14259_equation_0, values = (var_14033_cast, var_14215_cast))[name = tensor<string, []>("op_14259_cast")];
            tensor<string, []> var_14261_equation_0 = const()[name = tensor<string, []>("op_14261_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14261_cast = einsum(equation = var_14261_equation_0, values = (var_14033_cast, var_14216_cast))[name = tensor<string, []>("op_14261_cast")];
            tensor<string, []> var_14263_equation_0 = const()[name = tensor<string, []>("op_14263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14263_cast = einsum(equation = var_14263_equation_0, values = (var_14033_cast, var_14217_cast))[name = tensor<string, []>("op_14263_cast")];
            tensor<string, []> var_14265_equation_0 = const()[name = tensor<string, []>("op_14265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14265_cast = einsum(equation = var_14265_equation_0, values = (var_14033_cast, var_14218_cast))[name = tensor<string, []>("op_14265_cast")];
            tensor<string, []> var_14267_equation_0 = const()[name = tensor<string, []>("op_14267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14267_cast = einsum(equation = var_14267_equation_0, values = (var_14033_cast, var_14219_cast))[name = tensor<string, []>("op_14267_cast")];
            tensor<string, []> var_14269_equation_0 = const()[name = tensor<string, []>("op_14269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14269_cast = einsum(equation = var_14269_equation_0, values = (var_14037_cast, var_14220_cast))[name = tensor<string, []>("op_14269_cast")];
            tensor<string, []> var_14271_equation_0 = const()[name = tensor<string, []>("op_14271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14271_cast = einsum(equation = var_14271_equation_0, values = (var_14037_cast, var_14221_cast))[name = tensor<string, []>("op_14271_cast")];
            tensor<string, []> var_14273_equation_0 = const()[name = tensor<string, []>("op_14273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14273_cast = einsum(equation = var_14273_equation_0, values = (var_14037_cast, var_14222_cast))[name = tensor<string, []>("op_14273_cast")];
            tensor<string, []> var_14275_equation_0 = const()[name = tensor<string, []>("op_14275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14275_cast = einsum(equation = var_14275_equation_0, values = (var_14037_cast, var_14223_cast))[name = tensor<string, []>("op_14275_cast")];
            tensor<string, []> var_14277_equation_0 = const()[name = tensor<string, []>("op_14277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14277_cast = einsum(equation = var_14277_equation_0, values = (var_14037_cast, var_14224_cast))[name = tensor<string, []>("op_14277_cast")];
            tensor<string, []> var_14279_equation_0 = const()[name = tensor<string, []>("op_14279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14279_cast = einsum(equation = var_14279_equation_0, values = (var_14037_cast, var_14225_cast))[name = tensor<string, []>("op_14279_cast")];
            tensor<string, []> var_14281_equation_0 = const()[name = tensor<string, []>("op_14281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14281_cast = einsum(equation = var_14281_equation_0, values = (var_14037_cast, var_14226_cast))[name = tensor<string, []>("op_14281_cast")];
            tensor<string, []> var_14283_equation_0 = const()[name = tensor<string, []>("op_14283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14283_cast = einsum(equation = var_14283_equation_0, values = (var_14037_cast, var_14227_cast))[name = tensor<string, []>("op_14283_cast")];
            tensor<string, []> var_14285_equation_0 = const()[name = tensor<string, []>("op_14285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14285_cast = einsum(equation = var_14285_equation_0, values = (var_14041_cast, var_14228_cast))[name = tensor<string, []>("op_14285_cast")];
            tensor<string, []> var_14287_equation_0 = const()[name = tensor<string, []>("op_14287_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14287_cast = einsum(equation = var_14287_equation_0, values = (var_14041_cast, var_14229_cast))[name = tensor<string, []>("op_14287_cast")];
            tensor<string, []> var_14289_equation_0 = const()[name = tensor<string, []>("op_14289_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14289_cast = einsum(equation = var_14289_equation_0, values = (var_14041_cast, var_14230_cast))[name = tensor<string, []>("op_14289_cast")];
            tensor<string, []> var_14291_equation_0 = const()[name = tensor<string, []>("op_14291_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14291_cast = einsum(equation = var_14291_equation_0, values = (var_14041_cast, var_14231_cast))[name = tensor<string, []>("op_14291_cast")];
            tensor<string, []> var_14293_equation_0 = const()[name = tensor<string, []>("op_14293_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14293_cast = einsum(equation = var_14293_equation_0, values = (var_14041_cast, var_14232_cast))[name = tensor<string, []>("op_14293_cast")];
            tensor<string, []> var_14295_equation_0 = const()[name = tensor<string, []>("op_14295_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14295_cast = einsum(equation = var_14295_equation_0, values = (var_14041_cast, var_14233_cast))[name = tensor<string, []>("op_14295_cast")];
            tensor<string, []> var_14297_equation_0 = const()[name = tensor<string, []>("op_14297_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14297_cast = einsum(equation = var_14297_equation_0, values = (var_14041_cast, var_14234_cast))[name = tensor<string, []>("op_14297_cast")];
            tensor<string, []> var_14299_equation_0 = const()[name = tensor<string, []>("op_14299_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14299_cast = einsum(equation = var_14299_equation_0, values = (var_14041_cast, var_14235_cast))[name = tensor<string, []>("op_14299_cast")];
            tensor<string, []> var_14301_equation_0 = const()[name = tensor<string, []>("op_14301_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14301_cast = einsum(equation = var_14301_equation_0, values = (var_14045_cast, var_14236_cast))[name = tensor<string, []>("op_14301_cast")];
            tensor<string, []> var_14303_equation_0 = const()[name = tensor<string, []>("op_14303_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14303_cast = einsum(equation = var_14303_equation_0, values = (var_14045_cast, var_14237_cast))[name = tensor<string, []>("op_14303_cast")];
            tensor<string, []> var_14305_equation_0 = const()[name = tensor<string, []>("op_14305_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14305_cast = einsum(equation = var_14305_equation_0, values = (var_14045_cast, var_14238_cast))[name = tensor<string, []>("op_14305_cast")];
            tensor<string, []> var_14307_equation_0 = const()[name = tensor<string, []>("op_14307_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14307_cast = einsum(equation = var_14307_equation_0, values = (var_14045_cast, var_14239_cast))[name = tensor<string, []>("op_14307_cast")];
            tensor<string, []> var_14309_equation_0 = const()[name = tensor<string, []>("op_14309_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14309_cast = einsum(equation = var_14309_equation_0, values = (var_14045_cast, var_14240_cast))[name = tensor<string, []>("op_14309_cast")];
            tensor<string, []> var_14311_equation_0 = const()[name = tensor<string, []>("op_14311_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14311_cast = einsum(equation = var_14311_equation_0, values = (var_14045_cast, var_14241_cast))[name = tensor<string, []>("op_14311_cast")];
            tensor<string, []> var_14313_equation_0 = const()[name = tensor<string, []>("op_14313_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14313_cast = einsum(equation = var_14313_equation_0, values = (var_14045_cast, var_14242_cast))[name = tensor<string, []>("op_14313_cast")];
            tensor<string, []> var_14315_equation_0 = const()[name = tensor<string, []>("op_14315_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14315_cast = einsum(equation = var_14315_equation_0, values = (var_14045_cast, var_14243_cast))[name = tensor<string, []>("op_14315_cast")];
            tensor<string, []> var_14317_equation_0 = const()[name = tensor<string, []>("op_14317_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14317_cast = einsum(equation = var_14317_equation_0, values = (var_14049_cast, var_14244_cast))[name = tensor<string, []>("op_14317_cast")];
            tensor<string, []> var_14319_equation_0 = const()[name = tensor<string, []>("op_14319_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14319_cast = einsum(equation = var_14319_equation_0, values = (var_14049_cast, var_14245_cast))[name = tensor<string, []>("op_14319_cast")];
            tensor<string, []> var_14321_equation_0 = const()[name = tensor<string, []>("op_14321_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14321_cast = einsum(equation = var_14321_equation_0, values = (var_14049_cast, var_14246_cast))[name = tensor<string, []>("op_14321_cast")];
            tensor<string, []> var_14323_equation_0 = const()[name = tensor<string, []>("op_14323_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14323_cast = einsum(equation = var_14323_equation_0, values = (var_14049_cast, var_14247_cast))[name = tensor<string, []>("op_14323_cast")];
            tensor<string, []> var_14325_equation_0 = const()[name = tensor<string, []>("op_14325_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14325_cast = einsum(equation = var_14325_equation_0, values = (var_14049_cast, var_14248_cast))[name = tensor<string, []>("op_14325_cast")];
            tensor<string, []> var_14327_equation_0 = const()[name = tensor<string, []>("op_14327_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14327_cast = einsum(equation = var_14327_equation_0, values = (var_14049_cast, var_14249_cast))[name = tensor<string, []>("op_14327_cast")];
            tensor<string, []> var_14329_equation_0 = const()[name = tensor<string, []>("op_14329_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14329_cast = einsum(equation = var_14329_equation_0, values = (var_14049_cast, var_14250_cast))[name = tensor<string, []>("op_14329_cast")];
            tensor<string, []> var_14331_equation_0 = const()[name = tensor<string, []>("op_14331_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14331_cast = einsum(equation = var_14331_equation_0, values = (var_14049_cast, var_14251_cast))[name = tensor<string, []>("op_14331_cast")];
            tensor<bool, []> var_14333_interleave_0 = const()[name = tensor<string, []>("op_14333_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14333_cast = concat(axis = var_12782, interleave = var_14333_interleave_0, values = (var_14253_cast, var_14255_cast, var_14257_cast, var_14259_cast, var_14261_cast, var_14263_cast, var_14265_cast, var_14267_cast))[name = tensor<string, []>("op_14333_cast")];
            tensor<bool, []> var_14335_interleave_0 = const()[name = tensor<string, []>("op_14335_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14335_cast = concat(axis = var_12782, interleave = var_14335_interleave_0, values = (var_14269_cast, var_14271_cast, var_14273_cast, var_14275_cast, var_14277_cast, var_14279_cast, var_14281_cast, var_14283_cast))[name = tensor<string, []>("op_14335_cast")];
            tensor<bool, []> var_14337_interleave_0 = const()[name = tensor<string, []>("op_14337_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14337_cast = concat(axis = var_12782, interleave = var_14337_interleave_0, values = (var_14285_cast, var_14287_cast, var_14289_cast, var_14291_cast, var_14293_cast, var_14295_cast, var_14297_cast, var_14299_cast))[name = tensor<string, []>("op_14337_cast")];
            tensor<bool, []> var_14339_interleave_0 = const()[name = tensor<string, []>("op_14339_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14339_cast = concat(axis = var_12782, interleave = var_14339_interleave_0, values = (var_14301_cast, var_14303_cast, var_14305_cast, var_14307_cast, var_14309_cast, var_14311_cast, var_14313_cast, var_14315_cast))[name = tensor<string, []>("op_14339_cast")];
            tensor<bool, []> var_14341_interleave_0 = const()[name = tensor<string, []>("op_14341_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14341_cast = concat(axis = var_12782, interleave = var_14341_interleave_0, values = (var_14317_cast, var_14319_cast, var_14321_cast, var_14323_cast, var_14325_cast, var_14327_cast, var_14329_cast, var_14331_cast))[name = tensor<string, []>("op_14341_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_12807, interleave = input_491_interleave_0, values = (var_14333_cast, var_14335_cast, var_14337_cast, var_14339_cast, var_14341_cast))[name = tensor<string, []>("input_491_cast")];
            tensor<int32, [2]> var_14347 = const()[name = tensor<string, []>("op_14347"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14349 = const()[name = tensor<string, []>("op_14349"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14351_pad_type_0 = const()[name = tensor<string, []>("op_14351_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14351_pad_0 = const()[name = tensor<string, []>("op_14351_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214978624))), lut = tensor<fp16, [4]>([-0x1.7c8p-4, -0x1.aecp-6, 0x1.b54p-6, 0x1.7ccp-4]), 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, []>(215004288)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14351_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_14349, groups = var_12807, pad = var_14351_pad_0, pad_type = var_14351_pad_type_0, strides = var_14347, 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_14351_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_87_cast = add(x = var_14351_cast, y = inputs_85_cast)[name = tensor<string, []>("inputs_87_cast")];
            tensor<int32, [1]> var_14355 = const()[name = tensor<string, []>("op_14355"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_87_cast = reduce_mean(axes = var_14355, keep_dims = var_12802, 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_14359 = const()[name = tensor<string, []>("op_14359"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_14360_cast = reduce_mean(axes = var_14359, keep_dims = var_12802, x = zero_mean_sq_87_cast)[name = tensor<string, []>("op_14360_cast")];
            tensor<fp16, []> var_14361_to_fp16 = const()[name = tensor<string, []>("op_14361_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_14362_cast = add(x = var_14360_cast, y = var_14361_to_fp16)[name = tensor<string, []>("op_14362_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_14362_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_14366_to_fp16 = const()[name = tensor<string, []>("op_14366_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215004992)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14367_cast = add(x = out_87_cast, y = var_14366_to_fp16)[name = tensor<string, []>("op_14367_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14369_to_fp16 = const()[name = tensor<string, []>("op_14369_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215005696)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_309_cast = mul(x = var_14367_cast, y = var_14369_to_fp16)[name = tensor<string, []>("hidden_states_309_cast")];
            tensor<int32, [2]> var_14376 = const()[name = tensor<string, []>("op_14376"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14378 = const()[name = tensor<string, []>("op_14378"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215006400))), lut = tensor<fp16, [4]>([-0x1.af8p-4, -0x1.028p-5, 0x1.facp-6, 0x1.b14p-4]), 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_14378, groups = var_12807, pad = q_59_pad_0, pad_type = q_59_pad_type_0, strides = var_14376, 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_14382 = const()[name = tensor<string, []>("op_14382"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14384 = const()[name = tensor<string, []>("op_14384"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_117_pad_type_0 = const()[name = tensor<string, []>("k_117_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_117_pad_0 = const()[name = tensor<string, []>("k_117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215032064))), lut = tensor<fp16, [4]>([-0x1.54p-4, -0x1.7dp-6, 0x1.81cp-6, 0x1.55p-4]), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_117_cast = conv(dilations = var_14384, groups = var_12807, pad = k_117_pad_0, pad_type = k_117_pad_type_0, strides = var_14382, 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_14388 = const()[name = tensor<string, []>("op_14388"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14390 = const()[name = tensor<string, []>("op_14390"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_59_pad_type_0 = const()[name = tensor<string, []>("v_59_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_59_pad_0 = const()[name = tensor<string, []>("v_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215114048))), lut = tensor<fp16, [4]>([-0x1.a54p-6, -0x1.d2p-8, 0x1.df4p-8, 0x1.adp-6]), name = tensor<string, []>("up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_59_cast = conv(dilations = var_14390, groups = var_12807, pad = v_59_pad_0, pad_type = v_59_pad_type_0, strides = var_14388, 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_14394_begin_0 = const()[name = tensor<string, []>("op_14394_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14394_end_0 = const()[name = tensor<string, []>("op_14394_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14394_end_mask_0 = const()[name = tensor<string, []>("op_14394_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14394_cast = slice_by_index(begin = var_14394_begin_0, end = var_14394_end_0, end_mask = var_14394_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_14394_cast")];
            tensor<int32, [4]> var_14398_begin_0 = const()[name = tensor<string, []>("op_14398_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14398_end_0 = const()[name = tensor<string, []>("op_14398_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_14398_end_mask_0 = const()[name = tensor<string, []>("op_14398_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14398_cast = slice_by_index(begin = var_14398_begin_0, end = var_14398_end_0, end_mask = var_14398_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_14398_cast")];
            tensor<int32, [4]> var_14402_begin_0 = const()[name = tensor<string, []>("op_14402_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14402_end_0 = const()[name = tensor<string, []>("op_14402_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_14402_end_mask_0 = const()[name = tensor<string, []>("op_14402_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14402_cast = slice_by_index(begin = var_14402_begin_0, end = var_14402_end_0, end_mask = var_14402_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_14402_cast")];
            tensor<int32, [4]> var_14406_begin_0 = const()[name = tensor<string, []>("op_14406_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14406_end_0 = const()[name = tensor<string, []>("op_14406_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_14406_end_mask_0 = const()[name = tensor<string, []>("op_14406_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14406_cast = slice_by_index(begin = var_14406_begin_0, end = var_14406_end_0, end_mask = var_14406_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_14406_cast")];
            tensor<int32, [4]> var_14410_begin_0 = const()[name = tensor<string, []>("op_14410_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14410_end_0 = const()[name = tensor<string, []>("op_14410_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_14410_end_mask_0 = const()[name = tensor<string, []>("op_14410_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14410_cast = slice_by_index(begin = var_14410_begin_0, end = var_14410_end_0, end_mask = var_14410_end_mask_0, x = q_59_cast)[name = tensor<string, []>("op_14410_cast")];
            tensor<int32, [4]> var_14413_begin_0 = const()[name = tensor<string, []>("op_14413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14413_end_0 = const()[name = tensor<string, []>("op_14413_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14413_end_mask_0 = const()[name = tensor<string, []>("op_14413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14413_cast = slice_by_index(begin = var_14413_begin_0, end = var_14413_end_0, end_mask = var_14413_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14413_cast")];
            tensor<int32, [4]> var_14414_begin_0 = const()[name = tensor<string, []>("op_14414_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14414_end_0 = const()[name = tensor<string, []>("op_14414_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14414_end_mask_0 = const()[name = tensor<string, []>("op_14414_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14414_cast = slice_by_index(begin = var_14414_begin_0, end = var_14414_end_0, end_mask = var_14414_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14414_cast")];
            tensor<int32, [4]> var_14415_begin_0 = const()[name = tensor<string, []>("op_14415_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14415_end_0 = const()[name = tensor<string, []>("op_14415_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14415_end_mask_0 = const()[name = tensor<string, []>("op_14415_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14415_cast = slice_by_index(begin = var_14415_begin_0, end = var_14415_end_0, end_mask = var_14415_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14415_cast")];
            tensor<int32, [4]> var_14416_begin_0 = const()[name = tensor<string, []>("op_14416_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14416_end_0 = const()[name = tensor<string, []>("op_14416_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14416_end_mask_0 = const()[name = tensor<string, []>("op_14416_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14416_cast = slice_by_index(begin = var_14416_begin_0, end = var_14416_end_0, end_mask = var_14416_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14416_cast")];
            tensor<int32, [4]> var_14417_begin_0 = const()[name = tensor<string, []>("op_14417_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14417_end_0 = const()[name = tensor<string, []>("op_14417_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14417_end_mask_0 = const()[name = tensor<string, []>("op_14417_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14417_cast = slice_by_index(begin = var_14417_begin_0, end = var_14417_end_0, end_mask = var_14417_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14417_cast")];
            tensor<int32, [4]> var_14418_begin_0 = const()[name = tensor<string, []>("op_14418_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14418_end_0 = const()[name = tensor<string, []>("op_14418_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14418_end_mask_0 = const()[name = tensor<string, []>("op_14418_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14418_cast = slice_by_index(begin = var_14418_begin_0, end = var_14418_end_0, end_mask = var_14418_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14418_cast")];
            tensor<int32, [4]> var_14419_begin_0 = const()[name = tensor<string, []>("op_14419_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14419_end_0 = const()[name = tensor<string, []>("op_14419_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14419_end_mask_0 = const()[name = tensor<string, []>("op_14419_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14419_cast = slice_by_index(begin = var_14419_begin_0, end = var_14419_end_0, end_mask = var_14419_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14419_cast")];
            tensor<int32, [4]> var_14420_begin_0 = const()[name = tensor<string, []>("op_14420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14420_end_0 = const()[name = tensor<string, []>("op_14420_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14420_end_mask_0 = const()[name = tensor<string, []>("op_14420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14420_cast = slice_by_index(begin = var_14420_begin_0, end = var_14420_end_0, end_mask = var_14420_end_mask_0, x = var_14394_cast)[name = tensor<string, []>("op_14420_cast")];
            tensor<int32, [4]> var_14421_begin_0 = const()[name = tensor<string, []>("op_14421_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14421_end_0 = const()[name = tensor<string, []>("op_14421_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14421_end_mask_0 = const()[name = tensor<string, []>("op_14421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14421_cast = slice_by_index(begin = var_14421_begin_0, end = var_14421_end_0, end_mask = var_14421_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14421_cast")];
            tensor<int32, [4]> var_14422_begin_0 = const()[name = tensor<string, []>("op_14422_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14422_end_0 = const()[name = tensor<string, []>("op_14422_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14422_end_mask_0 = const()[name = tensor<string, []>("op_14422_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14422_cast = slice_by_index(begin = var_14422_begin_0, end = var_14422_end_0, end_mask = var_14422_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14422_cast")];
            tensor<int32, [4]> var_14423_begin_0 = const()[name = tensor<string, []>("op_14423_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14423_end_0 = const()[name = tensor<string, []>("op_14423_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14423_end_mask_0 = const()[name = tensor<string, []>("op_14423_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14423_cast = slice_by_index(begin = var_14423_begin_0, end = var_14423_end_0, end_mask = var_14423_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14423_cast")];
            tensor<int32, [4]> var_14424_begin_0 = const()[name = tensor<string, []>("op_14424_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14424_end_0 = const()[name = tensor<string, []>("op_14424_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14424_end_mask_0 = const()[name = tensor<string, []>("op_14424_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14424_cast = slice_by_index(begin = var_14424_begin_0, end = var_14424_end_0, end_mask = var_14424_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14424_cast")];
            tensor<int32, [4]> var_14425_begin_0 = const()[name = tensor<string, []>("op_14425_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14425_end_0 = const()[name = tensor<string, []>("op_14425_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14425_end_mask_0 = const()[name = tensor<string, []>("op_14425_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14425_cast = slice_by_index(begin = var_14425_begin_0, end = var_14425_end_0, end_mask = var_14425_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14425_cast")];
            tensor<int32, [4]> var_14426_begin_0 = const()[name = tensor<string, []>("op_14426_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14426_end_0 = const()[name = tensor<string, []>("op_14426_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14426_end_mask_0 = const()[name = tensor<string, []>("op_14426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14426_cast = slice_by_index(begin = var_14426_begin_0, end = var_14426_end_0, end_mask = var_14426_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14426_cast")];
            tensor<int32, [4]> var_14427_begin_0 = const()[name = tensor<string, []>("op_14427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14427_end_0 = const()[name = tensor<string, []>("op_14427_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14427_end_mask_0 = const()[name = tensor<string, []>("op_14427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14427_cast = slice_by_index(begin = var_14427_begin_0, end = var_14427_end_0, end_mask = var_14427_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14427_cast")];
            tensor<int32, [4]> var_14428_begin_0 = const()[name = tensor<string, []>("op_14428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14428_end_0 = const()[name = tensor<string, []>("op_14428_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14428_end_mask_0 = const()[name = tensor<string, []>("op_14428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14428_cast = slice_by_index(begin = var_14428_begin_0, end = var_14428_end_0, end_mask = var_14428_end_mask_0, x = var_14398_cast)[name = tensor<string, []>("op_14428_cast")];
            tensor<int32, [4]> var_14429_begin_0 = const()[name = tensor<string, []>("op_14429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14429_end_0 = const()[name = tensor<string, []>("op_14429_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14429_end_mask_0 = const()[name = tensor<string, []>("op_14429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14429_cast = slice_by_index(begin = var_14429_begin_0, end = var_14429_end_0, end_mask = var_14429_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14429_cast")];
            tensor<int32, [4]> var_14430_begin_0 = const()[name = tensor<string, []>("op_14430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14430_end_0 = const()[name = tensor<string, []>("op_14430_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14430_end_mask_0 = const()[name = tensor<string, []>("op_14430_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14430_cast = slice_by_index(begin = var_14430_begin_0, end = var_14430_end_0, end_mask = var_14430_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14430_cast")];
            tensor<int32, [4]> var_14431_begin_0 = const()[name = tensor<string, []>("op_14431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14431_end_0 = const()[name = tensor<string, []>("op_14431_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14431_end_mask_0 = const()[name = tensor<string, []>("op_14431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14431_cast = slice_by_index(begin = var_14431_begin_0, end = var_14431_end_0, end_mask = var_14431_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14431_cast")];
            tensor<int32, [4]> var_14432_begin_0 = const()[name = tensor<string, []>("op_14432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14432_end_0 = const()[name = tensor<string, []>("op_14432_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14432_end_mask_0 = const()[name = tensor<string, []>("op_14432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14432_cast = slice_by_index(begin = var_14432_begin_0, end = var_14432_end_0, end_mask = var_14432_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14432_cast")];
            tensor<int32, [4]> var_14433_begin_0 = const()[name = tensor<string, []>("op_14433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14433_end_0 = const()[name = tensor<string, []>("op_14433_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14433_end_mask_0 = const()[name = tensor<string, []>("op_14433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14433_cast = slice_by_index(begin = var_14433_begin_0, end = var_14433_end_0, end_mask = var_14433_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14433_cast")];
            tensor<int32, [4]> var_14434_begin_0 = const()[name = tensor<string, []>("op_14434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14434_end_0 = const()[name = tensor<string, []>("op_14434_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14434_end_mask_0 = const()[name = tensor<string, []>("op_14434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14434_cast = slice_by_index(begin = var_14434_begin_0, end = var_14434_end_0, end_mask = var_14434_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14434_cast")];
            tensor<int32, [4]> var_14435_begin_0 = const()[name = tensor<string, []>("op_14435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14435_end_0 = const()[name = tensor<string, []>("op_14435_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14435_end_mask_0 = const()[name = tensor<string, []>("op_14435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14435_cast = slice_by_index(begin = var_14435_begin_0, end = var_14435_end_0, end_mask = var_14435_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14435_cast")];
            tensor<int32, [4]> var_14436_begin_0 = const()[name = tensor<string, []>("op_14436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14436_end_0 = const()[name = tensor<string, []>("op_14436_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14436_end_mask_0 = const()[name = tensor<string, []>("op_14436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14436_cast = slice_by_index(begin = var_14436_begin_0, end = var_14436_end_0, end_mask = var_14436_end_mask_0, x = var_14402_cast)[name = tensor<string, []>("op_14436_cast")];
            tensor<int32, [4]> var_14437_begin_0 = const()[name = tensor<string, []>("op_14437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14437_end_0 = const()[name = tensor<string, []>("op_14437_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14437_end_mask_0 = const()[name = tensor<string, []>("op_14437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14437_cast = slice_by_index(begin = var_14437_begin_0, end = var_14437_end_0, end_mask = var_14437_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14437_cast")];
            tensor<int32, [4]> var_14438_begin_0 = const()[name = tensor<string, []>("op_14438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14438_end_0 = const()[name = tensor<string, []>("op_14438_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14438_end_mask_0 = const()[name = tensor<string, []>("op_14438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14438_cast = slice_by_index(begin = var_14438_begin_0, end = var_14438_end_0, end_mask = var_14438_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14438_cast")];
            tensor<int32, [4]> var_14439_begin_0 = const()[name = tensor<string, []>("op_14439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14439_end_0 = const()[name = tensor<string, []>("op_14439_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14439_end_mask_0 = const()[name = tensor<string, []>("op_14439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14439_cast = slice_by_index(begin = var_14439_begin_0, end = var_14439_end_0, end_mask = var_14439_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14439_cast")];
            tensor<int32, [4]> var_14440_begin_0 = const()[name = tensor<string, []>("op_14440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14440_end_0 = const()[name = tensor<string, []>("op_14440_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14440_end_mask_0 = const()[name = tensor<string, []>("op_14440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14440_cast = slice_by_index(begin = var_14440_begin_0, end = var_14440_end_0, end_mask = var_14440_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14440_cast")];
            tensor<int32, [4]> var_14441_begin_0 = const()[name = tensor<string, []>("op_14441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14441_end_0 = const()[name = tensor<string, []>("op_14441_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14441_end_mask_0 = const()[name = tensor<string, []>("op_14441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14441_cast = slice_by_index(begin = var_14441_begin_0, end = var_14441_end_0, end_mask = var_14441_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14441_cast")];
            tensor<int32, [4]> var_14442_begin_0 = const()[name = tensor<string, []>("op_14442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14442_end_0 = const()[name = tensor<string, []>("op_14442_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14442_end_mask_0 = const()[name = tensor<string, []>("op_14442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14442_cast = slice_by_index(begin = var_14442_begin_0, end = var_14442_end_0, end_mask = var_14442_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14442_cast")];
            tensor<int32, [4]> var_14443_begin_0 = const()[name = tensor<string, []>("op_14443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14443_end_0 = const()[name = tensor<string, []>("op_14443_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14443_end_mask_0 = const()[name = tensor<string, []>("op_14443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14443_cast = slice_by_index(begin = var_14443_begin_0, end = var_14443_end_0, end_mask = var_14443_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14443_cast")];
            tensor<int32, [4]> var_14444_begin_0 = const()[name = tensor<string, []>("op_14444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14444_end_0 = const()[name = tensor<string, []>("op_14444_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14444_end_mask_0 = const()[name = tensor<string, []>("op_14444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14444_cast = slice_by_index(begin = var_14444_begin_0, end = var_14444_end_0, end_mask = var_14444_end_mask_0, x = var_14406_cast)[name = tensor<string, []>("op_14444_cast")];
            tensor<int32, [4]> var_14445_begin_0 = const()[name = tensor<string, []>("op_14445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14445_end_0 = const()[name = tensor<string, []>("op_14445_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14445_end_mask_0 = const()[name = tensor<string, []>("op_14445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14445_cast = slice_by_index(begin = var_14445_begin_0, end = var_14445_end_0, end_mask = var_14445_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14445_cast")];
            tensor<int32, [4]> var_14446_begin_0 = const()[name = tensor<string, []>("op_14446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14446_end_0 = const()[name = tensor<string, []>("op_14446_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14446_end_mask_0 = const()[name = tensor<string, []>("op_14446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14446_cast = slice_by_index(begin = var_14446_begin_0, end = var_14446_end_0, end_mask = var_14446_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14446_cast")];
            tensor<int32, [4]> var_14447_begin_0 = const()[name = tensor<string, []>("op_14447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14447_end_0 = const()[name = tensor<string, []>("op_14447_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14447_end_mask_0 = const()[name = tensor<string, []>("op_14447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14447_cast = slice_by_index(begin = var_14447_begin_0, end = var_14447_end_0, end_mask = var_14447_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14447_cast")];
            tensor<int32, [4]> var_14448_begin_0 = const()[name = tensor<string, []>("op_14448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14448_end_0 = const()[name = tensor<string, []>("op_14448_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14448_end_mask_0 = const()[name = tensor<string, []>("op_14448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14448_cast = slice_by_index(begin = var_14448_begin_0, end = var_14448_end_0, end_mask = var_14448_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14448_cast")];
            tensor<int32, [4]> var_14449_begin_0 = const()[name = tensor<string, []>("op_14449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14449_end_0 = const()[name = tensor<string, []>("op_14449_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14449_end_mask_0 = const()[name = tensor<string, []>("op_14449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14449_cast = slice_by_index(begin = var_14449_begin_0, end = var_14449_end_0, end_mask = var_14449_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14449_cast")];
            tensor<int32, [4]> var_14450_begin_0 = const()[name = tensor<string, []>("op_14450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14450_end_0 = const()[name = tensor<string, []>("op_14450_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14450_end_mask_0 = const()[name = tensor<string, []>("op_14450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14450_cast = slice_by_index(begin = var_14450_begin_0, end = var_14450_end_0, end_mask = var_14450_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14450_cast")];
            tensor<int32, [4]> var_14451_begin_0 = const()[name = tensor<string, []>("op_14451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14451_end_0 = const()[name = tensor<string, []>("op_14451_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14451_end_mask_0 = const()[name = tensor<string, []>("op_14451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14451_cast = slice_by_index(begin = var_14451_begin_0, end = var_14451_end_0, end_mask = var_14451_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14451_cast")];
            tensor<int32, [4]> var_14452_begin_0 = const()[name = tensor<string, []>("op_14452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14452_end_0 = const()[name = tensor<string, []>("op_14452_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14452_end_mask_0 = const()[name = tensor<string, []>("op_14452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14452_cast = slice_by_index(begin = var_14452_begin_0, end = var_14452_end_0, end_mask = var_14452_end_mask_0, x = var_14410_cast)[name = tensor<string, []>("op_14452_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_14457_begin_0 = const()[name = tensor<string, []>("op_14457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14457_end_0 = const()[name = tensor<string, []>("op_14457_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_14457_end_mask_0 = const()[name = tensor<string, []>("op_14457_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, 64]> var_14457_cast = slice_by_index(begin = var_14457_begin_0, end = var_14457_end_0, end_mask = var_14457_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_14457_cast")];
            tensor<int32, [4]> var_14461_begin_0 = const()[name = tensor<string, []>("op_14461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_14461_end_0 = const()[name = tensor<string, []>("op_14461_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_14461_end_mask_0 = const()[name = tensor<string, []>("op_14461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14461_cast = slice_by_index(begin = var_14461_begin_0, end = var_14461_end_0, end_mask = var_14461_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_14461_cast")];
            tensor<int32, [4]> var_14465_begin_0 = const()[name = tensor<string, []>("op_14465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_14465_end_0 = const()[name = tensor<string, []>("op_14465_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_14465_end_mask_0 = const()[name = tensor<string, []>("op_14465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14465_cast = slice_by_index(begin = var_14465_begin_0, end = var_14465_end_0, end_mask = var_14465_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_14465_cast")];
            tensor<int32, [4]> var_14469_begin_0 = const()[name = tensor<string, []>("op_14469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_14469_end_0 = const()[name = tensor<string, []>("op_14469_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_14469_end_mask_0 = const()[name = tensor<string, []>("op_14469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14469_cast = slice_by_index(begin = var_14469_begin_0, end = var_14469_end_0, end_mask = var_14469_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_14469_cast")];
            tensor<int32, [4]> var_14473_begin_0 = const()[name = tensor<string, []>("op_14473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_14473_end_0 = const()[name = tensor<string, []>("op_14473_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_14473_end_mask_0 = const()[name = tensor<string, []>("op_14473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_14473_cast = slice_by_index(begin = var_14473_begin_0, end = var_14473_end_0, end_mask = var_14473_end_mask_0, x = transpose_2)[name = tensor<string, []>("op_14473_cast")];
            tensor<int32, [4]> var_14475_begin_0 = const()[name = tensor<string, []>("op_14475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14475_end_0 = const()[name = tensor<string, []>("op_14475_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_14475_end_mask_0 = const()[name = tensor<string, []>("op_14475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14475_cast = slice_by_index(begin = var_14475_begin_0, end = var_14475_end_0, end_mask = var_14475_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_14475_cast")];
            tensor<int32, [4]> var_14479_begin_0 = const()[name = tensor<string, []>("op_14479_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14479_end_0 = const()[name = tensor<string, []>("op_14479_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_14479_end_mask_0 = const()[name = tensor<string, []>("op_14479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14479_cast = slice_by_index(begin = var_14479_begin_0, end = var_14479_end_0, end_mask = var_14479_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_14479_cast")];
            tensor<int32, [4]> var_14483_begin_0 = const()[name = tensor<string, []>("op_14483_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14483_end_0 = const()[name = tensor<string, []>("op_14483_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_14483_end_mask_0 = const()[name = tensor<string, []>("op_14483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14483_cast = slice_by_index(begin = var_14483_begin_0, end = var_14483_end_0, end_mask = var_14483_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_14483_cast")];
            tensor<int32, [4]> var_14487_begin_0 = const()[name = tensor<string, []>("op_14487_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14487_end_0 = const()[name = tensor<string, []>("op_14487_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_14487_end_mask_0 = const()[name = tensor<string, []>("op_14487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14487_cast = slice_by_index(begin = var_14487_begin_0, end = var_14487_end_0, end_mask = var_14487_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_14487_cast")];
            tensor<int32, [4]> var_14491_begin_0 = const()[name = tensor<string, []>("op_14491_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14491_end_0 = const()[name = tensor<string, []>("op_14491_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_14491_end_mask_0 = const()[name = tensor<string, []>("op_14491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_14491_cast = slice_by_index(begin = var_14491_begin_0, end = var_14491_end_0, end_mask = var_14491_end_mask_0, x = v_59_cast)[name = tensor<string, []>("op_14491_cast")];
            tensor<string, []> var_14495_equation_0 = const()[name = tensor<string, []>("op_14495_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14495_cast = einsum(equation = var_14495_equation_0, values = (var_14457_cast, var_14413_cast))[name = tensor<string, []>("op_14495_cast")];
            tensor<fp16, []> var_14496_to_fp16 = const()[name = tensor<string, []>("op_14496_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_961_cast = mul(x = var_14495_cast, y = var_14496_to_fp16)[name = tensor<string, []>("aw_chunk_961_cast")];
            tensor<string, []> var_14499_equation_0 = const()[name = tensor<string, []>("op_14499_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14499_cast = einsum(equation = var_14499_equation_0, values = (var_14457_cast, var_14414_cast))[name = tensor<string, []>("op_14499_cast")];
            tensor<fp16, []> var_14500_to_fp16 = const()[name = tensor<string, []>("op_14500_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_963_cast = mul(x = var_14499_cast, y = var_14500_to_fp16)[name = tensor<string, []>("aw_chunk_963_cast")];
            tensor<string, []> var_14503_equation_0 = const()[name = tensor<string, []>("op_14503_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14503_cast = einsum(equation = var_14503_equation_0, values = (var_14457_cast, var_14415_cast))[name = tensor<string, []>("op_14503_cast")];
            tensor<fp16, []> var_14504_to_fp16 = const()[name = tensor<string, []>("op_14504_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_965_cast = mul(x = var_14503_cast, y = var_14504_to_fp16)[name = tensor<string, []>("aw_chunk_965_cast")];
            tensor<string, []> var_14507_equation_0 = const()[name = tensor<string, []>("op_14507_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14507_cast = einsum(equation = var_14507_equation_0, values = (var_14457_cast, var_14416_cast))[name = tensor<string, []>("op_14507_cast")];
            tensor<fp16, []> var_14508_to_fp16 = const()[name = tensor<string, []>("op_14508_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_967_cast = mul(x = var_14507_cast, y = var_14508_to_fp16)[name = tensor<string, []>("aw_chunk_967_cast")];
            tensor<string, []> var_14511_equation_0 = const()[name = tensor<string, []>("op_14511_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14511_cast = einsum(equation = var_14511_equation_0, values = (var_14457_cast, var_14417_cast))[name = tensor<string, []>("op_14511_cast")];
            tensor<fp16, []> var_14512_to_fp16 = const()[name = tensor<string, []>("op_14512_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_969_cast = mul(x = var_14511_cast, y = var_14512_to_fp16)[name = tensor<string, []>("aw_chunk_969_cast")];
            tensor<string, []> var_14515_equation_0 = const()[name = tensor<string, []>("op_14515_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14515_cast = einsum(equation = var_14515_equation_0, values = (var_14457_cast, var_14418_cast))[name = tensor<string, []>("op_14515_cast")];
            tensor<fp16, []> var_14516_to_fp16 = const()[name = tensor<string, []>("op_14516_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_971_cast = mul(x = var_14515_cast, y = var_14516_to_fp16)[name = tensor<string, []>("aw_chunk_971_cast")];
            tensor<string, []> var_14519_equation_0 = const()[name = tensor<string, []>("op_14519_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14519_cast = einsum(equation = var_14519_equation_0, values = (var_14457_cast, var_14419_cast))[name = tensor<string, []>("op_14519_cast")];
            tensor<fp16, []> var_14520_to_fp16 = const()[name = tensor<string, []>("op_14520_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_973_cast = mul(x = var_14519_cast, y = var_14520_to_fp16)[name = tensor<string, []>("aw_chunk_973_cast")];
            tensor<string, []> var_14523_equation_0 = const()[name = tensor<string, []>("op_14523_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14523_cast = einsum(equation = var_14523_equation_0, values = (var_14457_cast, var_14420_cast))[name = tensor<string, []>("op_14523_cast")];
            tensor<fp16, []> var_14524_to_fp16 = const()[name = tensor<string, []>("op_14524_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_975_cast = mul(x = var_14523_cast, y = var_14524_to_fp16)[name = tensor<string, []>("aw_chunk_975_cast")];
            tensor<string, []> var_14527_equation_0 = const()[name = tensor<string, []>("op_14527_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14527_cast = einsum(equation = var_14527_equation_0, values = (var_14461_cast, var_14421_cast))[name = tensor<string, []>("op_14527_cast")];
            tensor<fp16, []> var_14528_to_fp16 = const()[name = tensor<string, []>("op_14528_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_977_cast = mul(x = var_14527_cast, y = var_14528_to_fp16)[name = tensor<string, []>("aw_chunk_977_cast")];
            tensor<string, []> var_14531_equation_0 = const()[name = tensor<string, []>("op_14531_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14531_cast = einsum(equation = var_14531_equation_0, values = (var_14461_cast, var_14422_cast))[name = tensor<string, []>("op_14531_cast")];
            tensor<fp16, []> var_14532_to_fp16 = const()[name = tensor<string, []>("op_14532_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_979_cast = mul(x = var_14531_cast, y = var_14532_to_fp16)[name = tensor<string, []>("aw_chunk_979_cast")];
            tensor<string, []> var_14535_equation_0 = const()[name = tensor<string, []>("op_14535_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14535_cast = einsum(equation = var_14535_equation_0, values = (var_14461_cast, var_14423_cast))[name = tensor<string, []>("op_14535_cast")];
            tensor<fp16, []> var_14536_to_fp16 = const()[name = tensor<string, []>("op_14536_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_981_cast = mul(x = var_14535_cast, y = var_14536_to_fp16)[name = tensor<string, []>("aw_chunk_981_cast")];
            tensor<string, []> var_14539_equation_0 = const()[name = tensor<string, []>("op_14539_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14539_cast = einsum(equation = var_14539_equation_0, values = (var_14461_cast, var_14424_cast))[name = tensor<string, []>("op_14539_cast")];
            tensor<fp16, []> var_14540_to_fp16 = const()[name = tensor<string, []>("op_14540_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_983_cast = mul(x = var_14539_cast, y = var_14540_to_fp16)[name = tensor<string, []>("aw_chunk_983_cast")];
            tensor<string, []> var_14543_equation_0 = const()[name = tensor<string, []>("op_14543_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14543_cast = einsum(equation = var_14543_equation_0, values = (var_14461_cast, var_14425_cast))[name = tensor<string, []>("op_14543_cast")];
            tensor<fp16, []> var_14544_to_fp16 = const()[name = tensor<string, []>("op_14544_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_985_cast = mul(x = var_14543_cast, y = var_14544_to_fp16)[name = tensor<string, []>("aw_chunk_985_cast")];
            tensor<string, []> var_14547_equation_0 = const()[name = tensor<string, []>("op_14547_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14547_cast = einsum(equation = var_14547_equation_0, values = (var_14461_cast, var_14426_cast))[name = tensor<string, []>("op_14547_cast")];
            tensor<fp16, []> var_14548_to_fp16 = const()[name = tensor<string, []>("op_14548_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_987_cast = mul(x = var_14547_cast, y = var_14548_to_fp16)[name = tensor<string, []>("aw_chunk_987_cast")];
            tensor<string, []> var_14551_equation_0 = const()[name = tensor<string, []>("op_14551_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14551_cast = einsum(equation = var_14551_equation_0, values = (var_14461_cast, var_14427_cast))[name = tensor<string, []>("op_14551_cast")];
            tensor<fp16, []> var_14552_to_fp16 = const()[name = tensor<string, []>("op_14552_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_989_cast = mul(x = var_14551_cast, y = var_14552_to_fp16)[name = tensor<string, []>("aw_chunk_989_cast")];
            tensor<string, []> var_14555_equation_0 = const()[name = tensor<string, []>("op_14555_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14555_cast = einsum(equation = var_14555_equation_0, values = (var_14461_cast, var_14428_cast))[name = tensor<string, []>("op_14555_cast")];
            tensor<fp16, []> var_14556_to_fp16 = const()[name = tensor<string, []>("op_14556_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_991_cast = mul(x = var_14555_cast, y = var_14556_to_fp16)[name = tensor<string, []>("aw_chunk_991_cast")];
            tensor<string, []> var_14559_equation_0 = const()[name = tensor<string, []>("op_14559_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14559_cast = einsum(equation = var_14559_equation_0, values = (var_14465_cast, var_14429_cast))[name = tensor<string, []>("op_14559_cast")];
            tensor<fp16, []> var_14560_to_fp16 = const()[name = tensor<string, []>("op_14560_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_993_cast = mul(x = var_14559_cast, y = var_14560_to_fp16)[name = tensor<string, []>("aw_chunk_993_cast")];
            tensor<string, []> var_14563_equation_0 = const()[name = tensor<string, []>("op_14563_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14563_cast = einsum(equation = var_14563_equation_0, values = (var_14465_cast, var_14430_cast))[name = tensor<string, []>("op_14563_cast")];
            tensor<fp16, []> var_14564_to_fp16 = const()[name = tensor<string, []>("op_14564_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_995_cast = mul(x = var_14563_cast, y = var_14564_to_fp16)[name = tensor<string, []>("aw_chunk_995_cast")];
            tensor<string, []> var_14567_equation_0 = const()[name = tensor<string, []>("op_14567_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14567_cast = einsum(equation = var_14567_equation_0, values = (var_14465_cast, var_14431_cast))[name = tensor<string, []>("op_14567_cast")];
            tensor<fp16, []> var_14568_to_fp16 = const()[name = tensor<string, []>("op_14568_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_997_cast = mul(x = var_14567_cast, y = var_14568_to_fp16)[name = tensor<string, []>("aw_chunk_997_cast")];
            tensor<string, []> var_14571_equation_0 = const()[name = tensor<string, []>("op_14571_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14571_cast = einsum(equation = var_14571_equation_0, values = (var_14465_cast, var_14432_cast))[name = tensor<string, []>("op_14571_cast")];
            tensor<fp16, []> var_14572_to_fp16 = const()[name = tensor<string, []>("op_14572_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_999_cast = mul(x = var_14571_cast, y = var_14572_to_fp16)[name = tensor<string, []>("aw_chunk_999_cast")];
            tensor<string, []> var_14575_equation_0 = const()[name = tensor<string, []>("op_14575_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14575_cast = einsum(equation = var_14575_equation_0, values = (var_14465_cast, var_14433_cast))[name = tensor<string, []>("op_14575_cast")];
            tensor<fp16, []> var_14576_to_fp16 = const()[name = tensor<string, []>("op_14576_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1001_cast = mul(x = var_14575_cast, y = var_14576_to_fp16)[name = tensor<string, []>("aw_chunk_1001_cast")];
            tensor<string, []> var_14579_equation_0 = const()[name = tensor<string, []>("op_14579_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14579_cast = einsum(equation = var_14579_equation_0, values = (var_14465_cast, var_14434_cast))[name = tensor<string, []>("op_14579_cast")];
            tensor<fp16, []> var_14580_to_fp16 = const()[name = tensor<string, []>("op_14580_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1003_cast = mul(x = var_14579_cast, y = var_14580_to_fp16)[name = tensor<string, []>("aw_chunk_1003_cast")];
            tensor<string, []> var_14583_equation_0 = const()[name = tensor<string, []>("op_14583_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14583_cast = einsum(equation = var_14583_equation_0, values = (var_14465_cast, var_14435_cast))[name = tensor<string, []>("op_14583_cast")];
            tensor<fp16, []> var_14584_to_fp16 = const()[name = tensor<string, []>("op_14584_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1005_cast = mul(x = var_14583_cast, y = var_14584_to_fp16)[name = tensor<string, []>("aw_chunk_1005_cast")];
            tensor<string, []> var_14587_equation_0 = const()[name = tensor<string, []>("op_14587_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14587_cast = einsum(equation = var_14587_equation_0, values = (var_14465_cast, var_14436_cast))[name = tensor<string, []>("op_14587_cast")];
            tensor<fp16, []> var_14588_to_fp16 = const()[name = tensor<string, []>("op_14588_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1007_cast = mul(x = var_14587_cast, y = var_14588_to_fp16)[name = tensor<string, []>("aw_chunk_1007_cast")];
            tensor<string, []> var_14591_equation_0 = const()[name = tensor<string, []>("op_14591_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14591_cast = einsum(equation = var_14591_equation_0, values = (var_14469_cast, var_14437_cast))[name = tensor<string, []>("op_14591_cast")];
            tensor<fp16, []> var_14592_to_fp16 = const()[name = tensor<string, []>("op_14592_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1009_cast = mul(x = var_14591_cast, y = var_14592_to_fp16)[name = tensor<string, []>("aw_chunk_1009_cast")];
            tensor<string, []> var_14595_equation_0 = const()[name = tensor<string, []>("op_14595_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14595_cast = einsum(equation = var_14595_equation_0, values = (var_14469_cast, var_14438_cast))[name = tensor<string, []>("op_14595_cast")];
            tensor<fp16, []> var_14596_to_fp16 = const()[name = tensor<string, []>("op_14596_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1011_cast = mul(x = var_14595_cast, y = var_14596_to_fp16)[name = tensor<string, []>("aw_chunk_1011_cast")];
            tensor<string, []> var_14599_equation_0 = const()[name = tensor<string, []>("op_14599_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14599_cast = einsum(equation = var_14599_equation_0, values = (var_14469_cast, var_14439_cast))[name = tensor<string, []>("op_14599_cast")];
            tensor<fp16, []> var_14600_to_fp16 = const()[name = tensor<string, []>("op_14600_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1013_cast = mul(x = var_14599_cast, y = var_14600_to_fp16)[name = tensor<string, []>("aw_chunk_1013_cast")];
            tensor<string, []> var_14603_equation_0 = const()[name = tensor<string, []>("op_14603_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14603_cast = einsum(equation = var_14603_equation_0, values = (var_14469_cast, var_14440_cast))[name = tensor<string, []>("op_14603_cast")];
            tensor<fp16, []> var_14604_to_fp16 = const()[name = tensor<string, []>("op_14604_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1015_cast = mul(x = var_14603_cast, y = var_14604_to_fp16)[name = tensor<string, []>("aw_chunk_1015_cast")];
            tensor<string, []> var_14607_equation_0 = const()[name = tensor<string, []>("op_14607_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14607_cast = einsum(equation = var_14607_equation_0, values = (var_14469_cast, var_14441_cast))[name = tensor<string, []>("op_14607_cast")];
            tensor<fp16, []> var_14608_to_fp16 = const()[name = tensor<string, []>("op_14608_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1017_cast = mul(x = var_14607_cast, y = var_14608_to_fp16)[name = tensor<string, []>("aw_chunk_1017_cast")];
            tensor<string, []> var_14611_equation_0 = const()[name = tensor<string, []>("op_14611_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14611_cast = einsum(equation = var_14611_equation_0, values = (var_14469_cast, var_14442_cast))[name = tensor<string, []>("op_14611_cast")];
            tensor<fp16, []> var_14612_to_fp16 = const()[name = tensor<string, []>("op_14612_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1019_cast = mul(x = var_14611_cast, y = var_14612_to_fp16)[name = tensor<string, []>("aw_chunk_1019_cast")];
            tensor<string, []> var_14615_equation_0 = const()[name = tensor<string, []>("op_14615_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14615_cast = einsum(equation = var_14615_equation_0, values = (var_14469_cast, var_14443_cast))[name = tensor<string, []>("op_14615_cast")];
            tensor<fp16, []> var_14616_to_fp16 = const()[name = tensor<string, []>("op_14616_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1021_cast = mul(x = var_14615_cast, y = var_14616_to_fp16)[name = tensor<string, []>("aw_chunk_1021_cast")];
            tensor<string, []> var_14619_equation_0 = const()[name = tensor<string, []>("op_14619_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14619_cast = einsum(equation = var_14619_equation_0, values = (var_14469_cast, var_14444_cast))[name = tensor<string, []>("op_14619_cast")];
            tensor<fp16, []> var_14620_to_fp16 = const()[name = tensor<string, []>("op_14620_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1023_cast = mul(x = var_14619_cast, y = var_14620_to_fp16)[name = tensor<string, []>("aw_chunk_1023_cast")];
            tensor<string, []> var_14623_equation_0 = const()[name = tensor<string, []>("op_14623_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14623_cast = einsum(equation = var_14623_equation_0, values = (var_14473_cast, var_14445_cast))[name = tensor<string, []>("op_14623_cast")];
            tensor<fp16, []> var_14624_to_fp16 = const()[name = tensor<string, []>("op_14624_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1025_cast = mul(x = var_14623_cast, y = var_14624_to_fp16)[name = tensor<string, []>("aw_chunk_1025_cast")];
            tensor<string, []> var_14627_equation_0 = const()[name = tensor<string, []>("op_14627_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14627_cast = einsum(equation = var_14627_equation_0, values = (var_14473_cast, var_14446_cast))[name = tensor<string, []>("op_14627_cast")];
            tensor<fp16, []> var_14628_to_fp16 = const()[name = tensor<string, []>("op_14628_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1027_cast = mul(x = var_14627_cast, y = var_14628_to_fp16)[name = tensor<string, []>("aw_chunk_1027_cast")];
            tensor<string, []> var_14631_equation_0 = const()[name = tensor<string, []>("op_14631_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14631_cast = einsum(equation = var_14631_equation_0, values = (var_14473_cast, var_14447_cast))[name = tensor<string, []>("op_14631_cast")];
            tensor<fp16, []> var_14632_to_fp16 = const()[name = tensor<string, []>("op_14632_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1029_cast = mul(x = var_14631_cast, y = var_14632_to_fp16)[name = tensor<string, []>("aw_chunk_1029_cast")];
            tensor<string, []> var_14635_equation_0 = const()[name = tensor<string, []>("op_14635_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14635_cast = einsum(equation = var_14635_equation_0, values = (var_14473_cast, var_14448_cast))[name = tensor<string, []>("op_14635_cast")];
            tensor<fp16, []> var_14636_to_fp16 = const()[name = tensor<string, []>("op_14636_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1031_cast = mul(x = var_14635_cast, y = var_14636_to_fp16)[name = tensor<string, []>("aw_chunk_1031_cast")];
            tensor<string, []> var_14639_equation_0 = const()[name = tensor<string, []>("op_14639_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14639_cast = einsum(equation = var_14639_equation_0, values = (var_14473_cast, var_14449_cast))[name = tensor<string, []>("op_14639_cast")];
            tensor<fp16, []> var_14640_to_fp16 = const()[name = tensor<string, []>("op_14640_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1033_cast = mul(x = var_14639_cast, y = var_14640_to_fp16)[name = tensor<string, []>("aw_chunk_1033_cast")];
            tensor<string, []> var_14643_equation_0 = const()[name = tensor<string, []>("op_14643_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14643_cast = einsum(equation = var_14643_equation_0, values = (var_14473_cast, var_14450_cast))[name = tensor<string, []>("op_14643_cast")];
            tensor<fp16, []> var_14644_to_fp16 = const()[name = tensor<string, []>("op_14644_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1035_cast = mul(x = var_14643_cast, y = var_14644_to_fp16)[name = tensor<string, []>("aw_chunk_1035_cast")];
            tensor<string, []> var_14647_equation_0 = const()[name = tensor<string, []>("op_14647_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14647_cast = einsum(equation = var_14647_equation_0, values = (var_14473_cast, var_14451_cast))[name = tensor<string, []>("op_14647_cast")];
            tensor<fp16, []> var_14648_to_fp16 = const()[name = tensor<string, []>("op_14648_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1037_cast = mul(x = var_14647_cast, y = var_14648_to_fp16)[name = tensor<string, []>("aw_chunk_1037_cast")];
            tensor<string, []> var_14651_equation_0 = const()[name = tensor<string, []>("op_14651_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_14651_cast = einsum(equation = var_14651_equation_0, values = (var_14473_cast, var_14452_cast))[name = tensor<string, []>("op_14651_cast")];
            tensor<fp16, []> var_14652_to_fp16 = const()[name = tensor<string, []>("op_14652_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1039_cast = mul(x = var_14651_cast, y = var_14652_to_fp16)[name = tensor<string, []>("aw_chunk_1039_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14654_cast = softmax(axis = var_12807, x = aw_chunk_961_cast)[name = tensor<string, []>("op_14654_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14655_cast = softmax(axis = var_12807, x = aw_chunk_963_cast)[name = tensor<string, []>("op_14655_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14656_cast = softmax(axis = var_12807, x = aw_chunk_965_cast)[name = tensor<string, []>("op_14656_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14657_cast = softmax(axis = var_12807, x = aw_chunk_967_cast)[name = tensor<string, []>("op_14657_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14658_cast = softmax(axis = var_12807, x = aw_chunk_969_cast)[name = tensor<string, []>("op_14658_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14659_cast = softmax(axis = var_12807, x = aw_chunk_971_cast)[name = tensor<string, []>("op_14659_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14660_cast = softmax(axis = var_12807, x = aw_chunk_973_cast)[name = tensor<string, []>("op_14660_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14661_cast = softmax(axis = var_12807, x = aw_chunk_975_cast)[name = tensor<string, []>("op_14661_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14662_cast = softmax(axis = var_12807, x = aw_chunk_977_cast)[name = tensor<string, []>("op_14662_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14663_cast = softmax(axis = var_12807, x = aw_chunk_979_cast)[name = tensor<string, []>("op_14663_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14664_cast = softmax(axis = var_12807, x = aw_chunk_981_cast)[name = tensor<string, []>("op_14664_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14665_cast = softmax(axis = var_12807, x = aw_chunk_983_cast)[name = tensor<string, []>("op_14665_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14666_cast = softmax(axis = var_12807, x = aw_chunk_985_cast)[name = tensor<string, []>("op_14666_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14667_cast = softmax(axis = var_12807, x = aw_chunk_987_cast)[name = tensor<string, []>("op_14667_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14668_cast = softmax(axis = var_12807, x = aw_chunk_989_cast)[name = tensor<string, []>("op_14668_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14669_cast = softmax(axis = var_12807, x = aw_chunk_991_cast)[name = tensor<string, []>("op_14669_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14670_cast = softmax(axis = var_12807, x = aw_chunk_993_cast)[name = tensor<string, []>("op_14670_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14671_cast = softmax(axis = var_12807, x = aw_chunk_995_cast)[name = tensor<string, []>("op_14671_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14672_cast = softmax(axis = var_12807, x = aw_chunk_997_cast)[name = tensor<string, []>("op_14672_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14673_cast = softmax(axis = var_12807, x = aw_chunk_999_cast)[name = tensor<string, []>("op_14673_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14674_cast = softmax(axis = var_12807, x = aw_chunk_1001_cast)[name = tensor<string, []>("op_14674_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14675_cast = softmax(axis = var_12807, x = aw_chunk_1003_cast)[name = tensor<string, []>("op_14675_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14676_cast = softmax(axis = var_12807, x = aw_chunk_1005_cast)[name = tensor<string, []>("op_14676_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14677_cast = softmax(axis = var_12807, x = aw_chunk_1007_cast)[name = tensor<string, []>("op_14677_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14678_cast = softmax(axis = var_12807, x = aw_chunk_1009_cast)[name = tensor<string, []>("op_14678_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14679_cast = softmax(axis = var_12807, x = aw_chunk_1011_cast)[name = tensor<string, []>("op_14679_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14680_cast = softmax(axis = var_12807, x = aw_chunk_1013_cast)[name = tensor<string, []>("op_14680_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14681_cast = softmax(axis = var_12807, x = aw_chunk_1015_cast)[name = tensor<string, []>("op_14681_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14682_cast = softmax(axis = var_12807, x = aw_chunk_1017_cast)[name = tensor<string, []>("op_14682_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14683_cast = softmax(axis = var_12807, x = aw_chunk_1019_cast)[name = tensor<string, []>("op_14683_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14684_cast = softmax(axis = var_12807, x = aw_chunk_1021_cast)[name = tensor<string, []>("op_14684_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14685_cast = softmax(axis = var_12807, x = aw_chunk_1023_cast)[name = tensor<string, []>("op_14685_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14686_cast = softmax(axis = var_12807, x = aw_chunk_1025_cast)[name = tensor<string, []>("op_14686_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14687_cast = softmax(axis = var_12807, x = aw_chunk_1027_cast)[name = tensor<string, []>("op_14687_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14688_cast = softmax(axis = var_12807, x = aw_chunk_1029_cast)[name = tensor<string, []>("op_14688_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14689_cast = softmax(axis = var_12807, x = aw_chunk_1031_cast)[name = tensor<string, []>("op_14689_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14690_cast = softmax(axis = var_12807, x = aw_chunk_1033_cast)[name = tensor<string, []>("op_14690_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14691_cast = softmax(axis = var_12807, x = aw_chunk_1035_cast)[name = tensor<string, []>("op_14691_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14692_cast = softmax(axis = var_12807, x = aw_chunk_1037_cast)[name = tensor<string, []>("op_14692_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_14693_cast = softmax(axis = var_12807, x = aw_chunk_1039_cast)[name = tensor<string, []>("op_14693_cast")];
            tensor<string, []> var_14695_equation_0 = const()[name = tensor<string, []>("op_14695_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14695_cast = einsum(equation = var_14695_equation_0, values = (var_14475_cast, var_14654_cast))[name = tensor<string, []>("op_14695_cast")];
            tensor<string, []> var_14697_equation_0 = const()[name = tensor<string, []>("op_14697_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14697_cast = einsum(equation = var_14697_equation_0, values = (var_14475_cast, var_14655_cast))[name = tensor<string, []>("op_14697_cast")];
            tensor<string, []> var_14699_equation_0 = const()[name = tensor<string, []>("op_14699_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14699_cast = einsum(equation = var_14699_equation_0, values = (var_14475_cast, var_14656_cast))[name = tensor<string, []>("op_14699_cast")];
            tensor<string, []> var_14701_equation_0 = const()[name = tensor<string, []>("op_14701_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14701_cast = einsum(equation = var_14701_equation_0, values = (var_14475_cast, var_14657_cast))[name = tensor<string, []>("op_14701_cast")];
            tensor<string, []> var_14703_equation_0 = const()[name = tensor<string, []>("op_14703_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14703_cast = einsum(equation = var_14703_equation_0, values = (var_14475_cast, var_14658_cast))[name = tensor<string, []>("op_14703_cast")];
            tensor<string, []> var_14705_equation_0 = const()[name = tensor<string, []>("op_14705_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14705_cast = einsum(equation = var_14705_equation_0, values = (var_14475_cast, var_14659_cast))[name = tensor<string, []>("op_14705_cast")];
            tensor<string, []> var_14707_equation_0 = const()[name = tensor<string, []>("op_14707_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14707_cast = einsum(equation = var_14707_equation_0, values = (var_14475_cast, var_14660_cast))[name = tensor<string, []>("op_14707_cast")];
            tensor<string, []> var_14709_equation_0 = const()[name = tensor<string, []>("op_14709_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14709_cast = einsum(equation = var_14709_equation_0, values = (var_14475_cast, var_14661_cast))[name = tensor<string, []>("op_14709_cast")];
            tensor<string, []> var_14711_equation_0 = const()[name = tensor<string, []>("op_14711_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14711_cast = einsum(equation = var_14711_equation_0, values = (var_14479_cast, var_14662_cast))[name = tensor<string, []>("op_14711_cast")];
            tensor<string, []> var_14713_equation_0 = const()[name = tensor<string, []>("op_14713_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14713_cast = einsum(equation = var_14713_equation_0, values = (var_14479_cast, var_14663_cast))[name = tensor<string, []>("op_14713_cast")];
            tensor<string, []> var_14715_equation_0 = const()[name = tensor<string, []>("op_14715_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14715_cast = einsum(equation = var_14715_equation_0, values = (var_14479_cast, var_14664_cast))[name = tensor<string, []>("op_14715_cast")];
            tensor<string, []> var_14717_equation_0 = const()[name = tensor<string, []>("op_14717_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14717_cast = einsum(equation = var_14717_equation_0, values = (var_14479_cast, var_14665_cast))[name = tensor<string, []>("op_14717_cast")];
            tensor<string, []> var_14719_equation_0 = const()[name = tensor<string, []>("op_14719_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14719_cast = einsum(equation = var_14719_equation_0, values = (var_14479_cast, var_14666_cast))[name = tensor<string, []>("op_14719_cast")];
            tensor<string, []> var_14721_equation_0 = const()[name = tensor<string, []>("op_14721_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14721_cast = einsum(equation = var_14721_equation_0, values = (var_14479_cast, var_14667_cast))[name = tensor<string, []>("op_14721_cast")];
            tensor<string, []> var_14723_equation_0 = const()[name = tensor<string, []>("op_14723_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14723_cast = einsum(equation = var_14723_equation_0, values = (var_14479_cast, var_14668_cast))[name = tensor<string, []>("op_14723_cast")];
            tensor<string, []> var_14725_equation_0 = const()[name = tensor<string, []>("op_14725_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14725_cast = einsum(equation = var_14725_equation_0, values = (var_14479_cast, var_14669_cast))[name = tensor<string, []>("op_14725_cast")];
            tensor<string, []> var_14727_equation_0 = const()[name = tensor<string, []>("op_14727_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14727_cast = einsum(equation = var_14727_equation_0, values = (var_14483_cast, var_14670_cast))[name = tensor<string, []>("op_14727_cast")];
            tensor<string, []> var_14729_equation_0 = const()[name = tensor<string, []>("op_14729_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14729_cast = einsum(equation = var_14729_equation_0, values = (var_14483_cast, var_14671_cast))[name = tensor<string, []>("op_14729_cast")];
            tensor<string, []> var_14731_equation_0 = const()[name = tensor<string, []>("op_14731_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14731_cast = einsum(equation = var_14731_equation_0, values = (var_14483_cast, var_14672_cast))[name = tensor<string, []>("op_14731_cast")];
            tensor<string, []> var_14733_equation_0 = const()[name = tensor<string, []>("op_14733_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14733_cast = einsum(equation = var_14733_equation_0, values = (var_14483_cast, var_14673_cast))[name = tensor<string, []>("op_14733_cast")];
            tensor<string, []> var_14735_equation_0 = const()[name = tensor<string, []>("op_14735_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14735_cast = einsum(equation = var_14735_equation_0, values = (var_14483_cast, var_14674_cast))[name = tensor<string, []>("op_14735_cast")];
            tensor<string, []> var_14737_equation_0 = const()[name = tensor<string, []>("op_14737_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14737_cast = einsum(equation = var_14737_equation_0, values = (var_14483_cast, var_14675_cast))[name = tensor<string, []>("op_14737_cast")];
            tensor<string, []> var_14739_equation_0 = const()[name = tensor<string, []>("op_14739_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14739_cast = einsum(equation = var_14739_equation_0, values = (var_14483_cast, var_14676_cast))[name = tensor<string, []>("op_14739_cast")];
            tensor<string, []> var_14741_equation_0 = const()[name = tensor<string, []>("op_14741_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14741_cast = einsum(equation = var_14741_equation_0, values = (var_14483_cast, var_14677_cast))[name = tensor<string, []>("op_14741_cast")];
            tensor<string, []> var_14743_equation_0 = const()[name = tensor<string, []>("op_14743_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14743_cast = einsum(equation = var_14743_equation_0, values = (var_14487_cast, var_14678_cast))[name = tensor<string, []>("op_14743_cast")];
            tensor<string, []> var_14745_equation_0 = const()[name = tensor<string, []>("op_14745_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14745_cast = einsum(equation = var_14745_equation_0, values = (var_14487_cast, var_14679_cast))[name = tensor<string, []>("op_14745_cast")];
            tensor<string, []> var_14747_equation_0 = const()[name = tensor<string, []>("op_14747_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14747_cast = einsum(equation = var_14747_equation_0, values = (var_14487_cast, var_14680_cast))[name = tensor<string, []>("op_14747_cast")];
            tensor<string, []> var_14749_equation_0 = const()[name = tensor<string, []>("op_14749_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14749_cast = einsum(equation = var_14749_equation_0, values = (var_14487_cast, var_14681_cast))[name = tensor<string, []>("op_14749_cast")];
            tensor<string, []> var_14751_equation_0 = const()[name = tensor<string, []>("op_14751_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14751_cast = einsum(equation = var_14751_equation_0, values = (var_14487_cast, var_14682_cast))[name = tensor<string, []>("op_14751_cast")];
            tensor<string, []> var_14753_equation_0 = const()[name = tensor<string, []>("op_14753_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14753_cast = einsum(equation = var_14753_equation_0, values = (var_14487_cast, var_14683_cast))[name = tensor<string, []>("op_14753_cast")];
            tensor<string, []> var_14755_equation_0 = const()[name = tensor<string, []>("op_14755_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14755_cast = einsum(equation = var_14755_equation_0, values = (var_14487_cast, var_14684_cast))[name = tensor<string, []>("op_14755_cast")];
            tensor<string, []> var_14757_equation_0 = const()[name = tensor<string, []>("op_14757_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14757_cast = einsum(equation = var_14757_equation_0, values = (var_14487_cast, var_14685_cast))[name = tensor<string, []>("op_14757_cast")];
            tensor<string, []> var_14759_equation_0 = const()[name = tensor<string, []>("op_14759_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14759_cast = einsum(equation = var_14759_equation_0, values = (var_14491_cast, var_14686_cast))[name = tensor<string, []>("op_14759_cast")];
            tensor<string, []> var_14761_equation_0 = const()[name = tensor<string, []>("op_14761_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14761_cast = einsum(equation = var_14761_equation_0, values = (var_14491_cast, var_14687_cast))[name = tensor<string, []>("op_14761_cast")];
            tensor<string, []> var_14763_equation_0 = const()[name = tensor<string, []>("op_14763_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14763_cast = einsum(equation = var_14763_equation_0, values = (var_14491_cast, var_14688_cast))[name = tensor<string, []>("op_14763_cast")];
            tensor<string, []> var_14765_equation_0 = const()[name = tensor<string, []>("op_14765_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14765_cast = einsum(equation = var_14765_equation_0, values = (var_14491_cast, var_14689_cast))[name = tensor<string, []>("op_14765_cast")];
            tensor<string, []> var_14767_equation_0 = const()[name = tensor<string, []>("op_14767_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14767_cast = einsum(equation = var_14767_equation_0, values = (var_14491_cast, var_14690_cast))[name = tensor<string, []>("op_14767_cast")];
            tensor<string, []> var_14769_equation_0 = const()[name = tensor<string, []>("op_14769_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14769_cast = einsum(equation = var_14769_equation_0, values = (var_14491_cast, var_14691_cast))[name = tensor<string, []>("op_14769_cast")];
            tensor<string, []> var_14771_equation_0 = const()[name = tensor<string, []>("op_14771_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14771_cast = einsum(equation = var_14771_equation_0, values = (var_14491_cast, var_14692_cast))[name = tensor<string, []>("op_14771_cast")];
            tensor<string, []> var_14773_equation_0 = const()[name = tensor<string, []>("op_14773_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_14773_cast = einsum(equation = var_14773_equation_0, values = (var_14491_cast, var_14693_cast))[name = tensor<string, []>("op_14773_cast")];
            tensor<bool, []> var_14775_interleave_0 = const()[name = tensor<string, []>("op_14775_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14775_cast = concat(axis = var_12782, interleave = var_14775_interleave_0, values = (var_14695_cast, var_14697_cast, var_14699_cast, var_14701_cast, var_14703_cast, var_14705_cast, var_14707_cast, var_14709_cast))[name = tensor<string, []>("op_14775_cast")];
            tensor<bool, []> var_14777_interleave_0 = const()[name = tensor<string, []>("op_14777_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14777_cast = concat(axis = var_12782, interleave = var_14777_interleave_0, values = (var_14711_cast, var_14713_cast, var_14715_cast, var_14717_cast, var_14719_cast, var_14721_cast, var_14723_cast, var_14725_cast))[name = tensor<string, []>("op_14777_cast")];
            tensor<bool, []> var_14779_interleave_0 = const()[name = tensor<string, []>("op_14779_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14779_cast = concat(axis = var_12782, interleave = var_14779_interleave_0, values = (var_14727_cast, var_14729_cast, var_14731_cast, var_14733_cast, var_14735_cast, var_14737_cast, var_14739_cast, var_14741_cast))[name = tensor<string, []>("op_14779_cast")];
            tensor<bool, []> var_14781_interleave_0 = const()[name = tensor<string, []>("op_14781_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14781_cast = concat(axis = var_12782, interleave = var_14781_interleave_0, values = (var_14743_cast, var_14745_cast, var_14747_cast, var_14749_cast, var_14751_cast, var_14753_cast, var_14755_cast, var_14757_cast))[name = tensor<string, []>("op_14781_cast")];
            tensor<bool, []> var_14783_interleave_0 = const()[name = tensor<string, []>("op_14783_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_14783_cast = concat(axis = var_12782, interleave = var_14783_interleave_0, values = (var_14759_cast, var_14761_cast, var_14763_cast, var_14765_cast, var_14767_cast, var_14769_cast, var_14771_cast, var_14773_cast))[name = tensor<string, []>("op_14783_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_12807, interleave = input_493_interleave_0, values = (var_14775_cast, var_14777_cast, var_14779_cast, var_14781_cast, var_14783_cast))[name = tensor<string, []>("input_493_cast")];
            tensor<int32, [2]> var_14789 = const()[name = tensor<string, []>("op_14789"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14791 = const()[name = tensor<string, []>("op_14791"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14793_pad_type_0 = const()[name = tensor<string, []>("op_14793_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14793_pad_0 = const()[name = tensor<string, []>("op_14793_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215196032))), lut = tensor<fp16, [4]>([-0x1.8bcp-6, -0x1.38cp-9, 0x1.38p-9, 0x1.7bp-6]), 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, []>(215221696)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14793_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_14791, groups = var_12807, pad = var_14793_pad_0, pad_type = var_14793_pad_type_0, strides = var_14789, 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_14793_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_89_cast = add(x = var_14793_cast, y = inputs_87_cast)[name = tensor<string, []>("inputs_89_cast")];
            tensor<int32, [1]> var_14797 = const()[name = tensor<string, []>("op_14797"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_89_cast = reduce_mean(axes = var_14797, keep_dims = var_12802, 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_14801 = const()[name = tensor<string, []>("op_14801"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_14802_cast = reduce_mean(axes = var_14801, keep_dims = var_12802, x = zero_mean_sq_89_cast)[name = tensor<string, []>("op_14802_cast")];
            tensor<fp16, []> var_14803_to_fp16 = const()[name = tensor<string, []>("op_14803_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_14804_cast = add(x = var_14802_cast, y = var_14803_to_fp16)[name = tensor<string, []>("op_14804_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_14804_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_14808_to_fp16 = const()[name = tensor<string, []>("op_14808_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215222400)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14809_cast = add(x = out_89_cast, y = var_14808_to_fp16)[name = tensor<string, []>("op_14809_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14811_to_fp16 = const()[name = tensor<string, []>("op_14811_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215223104)))];
            tensor<fp16, [2, 320, 1, 4096]> input_495_cast = mul(x = var_14809_cast, y = var_14811_to_fp16)[name = tensor<string, []>("input_495_cast")];
            tensor<int32, [2]> var_14819 = const()[name = tensor<string, []>("op_14819"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14821 = const()[name = tensor<string, []>("op_14821"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14823_pad_type_0 = const()[name = tensor<string, []>("op_14823_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14823_pad_0 = const()[name = tensor<string, []>("op_14823_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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215223808))), lut = tensor<fp16, [4]>([-0x1.7bcp-4, -0x1.bc4p-6, 0x1.b9p-6, 0x1.788p-4]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215428672))), lut = tensor<fp16, [4]>([0x1.6d4p-4, -0x1.89cp-6, -0x1.454p-2, 0x1.068p-5]), 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_14823_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_14821, groups = var_12807, pad = var_14823_pad_0, pad_type = var_14823_pad_type_0, strides = var_14819, 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_14823_cast")];
            tensor<int32, [2]> var_14824_split_sizes_0 = const()[name = tensor<string, []>("op_14824_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_14824_axis_0 = const()[name = tensor<string, []>("op_14824_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_14824_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_14824_cast_1 = split(axis = var_14824_axis_0, split_sizes = var_14824_split_sizes_0, x = var_14823_cast)[name = tensor<string, []>("op_14824_cast")];
            tensor<string, []> var_14826_mode_0 = const()[name = tensor<string, []>("op_14826_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_14826_cast = gelu(mode = var_14826_mode_0, x = var_14824_cast_1)[name = tensor<string, []>("op_14826_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_497_cast = mul(x = var_14824_cast_0, y = var_14826_cast)[name = tensor<string, []>("input_497_cast")];
            tensor<int32, [2]> var_14830 = const()[name = tensor<string, []>("op_14830"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14832 = const()[name = tensor<string, []>("op_14832"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_14834_pad_type_0 = const()[name = tensor<string, []>("op_14834_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_14834_pad_0 = const()[name = tensor<string, []>("op_14834_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215429376))), lut = tensor<fp16, [4]>([-0x1.97cp-4, -0x1.e3cp-6, 0x1.e24p-6, 0x1.97cp-4]), 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, []>(215531840)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14834_cast = conv(bias = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_14832, groups = var_12807, pad = var_14834_pad_0, pad_type = var_14834_pad_type_0, strides = var_14830, weight = up_blocks_3_attentions_1_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_497_cast)[name = tensor<string, []>("op_14834_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_313_cast = add(x = var_14834_cast, y = inputs_89_cast)[name = tensor<string, []>("hidden_states_313_cast")];
            tensor<int32, [4]> var_14836 = const()[name = tensor<string, []>("op_14836"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_499_cast = reshape(shape = var_14836, x = hidden_states_313_cast)[name = tensor<string, []>("input_499_cast")];
            tensor<int32, [2]> var_14840 = const()[name = tensor<string, []>("op_14840"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14842 = const()[name = tensor<string, []>("op_14842"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215532544))), lut = tensor<fp16, [4]>([-0x1.0f8p-3, -0x1.3f8p-5, 0x1.474p-5, 0x1.10cp-3]), 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, []>(215558208)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_315_cast = conv(bias = up_blocks_3_attentions_1_proj_out_bias_to_fp16, dilations = var_14842, groups = var_12807, pad = hidden_states_315_pad_0, pad_type = hidden_states_315_pad_type_0, strides = var_14840, 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_12807, 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, []>(215558912)))];
            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, []>(215560256)))];
            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_14860 = const()[name = tensor<string, []>("op_14860"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14862 = const()[name = tensor<string, []>("op_14862"), 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, [460800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215561600))), lut = tensor<fp16, [4]>([-0x1.93p-4, -0x1.76p-6, 0x1.928p-6, 0x1.9ccp-4]), 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, []>(216022464)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_319_cast = conv(bias = up_blocks_3_resnets_2_conv1_bias_to_fp16, dilations = var_14862, groups = var_12807, pad = hidden_states_319_pad_0, pad_type = hidden_states_319_pad_type_0, strides = var_14860, 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_14868 = const()[name = tensor<string, []>("op_14868"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14870 = const()[name = tensor<string, []>("op_14870"), 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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216023168))), lut = tensor<fp16, [4]>([-0x1.f88p-5, -0x1.8ap-9, 0x1.678p-9, 0x1.91p-5]), 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, []>(216125632)))];
            tensor<fp16, [2, 320, 1, 1]> temb_cast = conv(bias = up_blocks_3_resnets_2_time_emb_proj_bias_to_fp16, dilations = var_14870, groups = var_12807, pad = temb_pad_0, pad_type = temb_pad_type_0, strides = var_14868, 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, []>(216126336)))];
            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, []>(216127040)))];
            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_14880 = const()[name = tensor<string, []>("op_14880"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14882 = const()[name = tensor<string, []>("op_14882"), 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, [230400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216127744))), lut = tensor<fp16, [4]>([-0x1.89cp-4, -0x1.97cp-6, 0x1.878p-6, 0x1.828p-4]), 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, []>(216358208)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_321_cast = conv(bias = up_blocks_3_resnets_2_conv2_bias_to_fp16, dilations = var_14882, groups = var_12807, pad = hidden_states_321_pad_0, pad_type = hidden_states_321_pad_type_0, strides = var_14880, 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_14887 = const()[name = tensor<string, []>("op_14887"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14889 = const()[name = tensor<string, []>("op_14889"), 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, [51200]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216358912))), lut = tensor<fp16, [4]>([-0x1.2c4p-4, -0x1.1f8p-6, 0x1.324p-6, 0x1.338p-4]), 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, []>(216410176)))];
            tensor<fp16, [2, 320, 64, 64]> x_cast = conv(bias = up_blocks_3_resnets_2_conv_shortcut_bias_to_fp16, dilations = var_14889, groups = var_12807, pad = x_pad_0, pad_type = x_pad_type_0, strides = var_14887, 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, []>(216410880)))];
            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, []>(216411584)))];
            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_14909 = const()[name = tensor<string, []>("op_14909"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14911 = const()[name = tensor<string, []>("op_14911"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216412288))), lut = tensor<fp16, [4]>([-0x1.8d8p-4, -0x1.ddp-6, 0x1.d6p-6, 0x1.88p-4]), 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, []>(216437952)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_325_cast = conv(bias = up_blocks_3_attentions_2_proj_in_bias_to_fp16, dilations = var_14911, groups = var_12807, pad = hidden_states_325_pad_0, pad_type = hidden_states_325_pad_type_0, strides = var_14909, 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_14916 = const()[name = tensor<string, []>("op_14916"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<fp16, [2, 320, 1, 4096]> inputs_91_cast = reshape(shape = var_14916, x = hidden_states_325_cast)[name = tensor<string, []>("inputs_91_cast")];
            tensor<int32, [1]> var_14926 = const()[name = tensor<string, []>("op_14926"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_91_cast = reduce_mean(axes = var_14926, keep_dims = var_12802, 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_14930 = const()[name = tensor<string, []>("op_14930"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_14931_cast = reduce_mean(axes = var_14930, keep_dims = var_12802, x = zero_mean_sq_91_cast)[name = tensor<string, []>("op_14931_cast")];
            tensor<fp16, []> var_14932_to_fp16 = const()[name = tensor<string, []>("op_14932_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_14933_cast = add(x = var_14931_cast, y = var_14932_to_fp16)[name = tensor<string, []>("op_14933_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_14933_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_14937_to_fp16 = const()[name = tensor<string, []>("op_14937_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216438656)))];
            tensor<fp16, [2, 320, 1, 4096]> var_14938_cast = add(x = out_91_cast, y = var_14937_to_fp16)[name = tensor<string, []>("op_14938_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_14940_to_fp16 = const()[name = tensor<string, []>("op_14940_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216439360)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_327_cast = mul(x = var_14938_cast, y = var_14940_to_fp16)[name = tensor<string, []>("hidden_states_327_cast")];
            tensor<int32, [2]> var_14947 = const()[name = tensor<string, []>("op_14947"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14949 = const()[name = tensor<string, []>("op_14949"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216440064))), lut = tensor<fp16, [4]>([-0x1.748p-3, -0x1.b18p-5, 0x1.a68p-5, 0x1.74p-3]), 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_14949, groups = var_12807, pad = q_61_pad_0, pad_type = q_61_pad_type_0, strides = var_14947, 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_14953 = const()[name = tensor<string, []>("op_14953"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14955 = const()[name = tensor<string, []>("op_14955"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216465728))), lut = tensor<fp16, [4]>([-0x1.65p-3, -0x1.7d4p-5, 0x1.87p-5, 0x1.688p-3]), 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_14955, groups = var_12807, pad = k_121_pad_0, pad_type = k_121_pad_type_0, strides = var_14953, 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_14959 = const()[name = tensor<string, []>("op_14959"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_14961 = const()[name = tensor<string, []>("op_14961"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216491392))), lut = tensor<fp16, [4]>([-0x1.7p-4, -0x1.9b8p-6, 0x1.9b4p-6, 0x1.6fp-4]), 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_14961, groups = var_12807, pad = v_61_pad_0, pad_type = v_61_pad_type_0, strides = var_14959, 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_14965_begin_0 = const()[name = tensor<string, []>("op_14965_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14965_end_0 = const()[name = tensor<string, []>("op_14965_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14965_end_mask_0 = const()[name = tensor<string, []>("op_14965_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14965_cast = slice_by_index(begin = var_14965_begin_0, end = var_14965_end_0, end_mask = var_14965_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_14965_cast")];
            tensor<int32, [4]> var_14969_begin_0 = const()[name = tensor<string, []>("op_14969_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_14969_end_0 = const()[name = tensor<string, []>("op_14969_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_14969_end_mask_0 = const()[name = tensor<string, []>("op_14969_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14969_cast = slice_by_index(begin = var_14969_begin_0, end = var_14969_end_0, end_mask = var_14969_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_14969_cast")];
            tensor<int32, [4]> var_14973_begin_0 = const()[name = tensor<string, []>("op_14973_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_14973_end_0 = const()[name = tensor<string, []>("op_14973_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_14973_end_mask_0 = const()[name = tensor<string, []>("op_14973_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14973_cast = slice_by_index(begin = var_14973_begin_0, end = var_14973_end_0, end_mask = var_14973_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_14973_cast")];
            tensor<int32, [4]> var_14977_begin_0 = const()[name = tensor<string, []>("op_14977_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_14977_end_0 = const()[name = tensor<string, []>("op_14977_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_14977_end_mask_0 = const()[name = tensor<string, []>("op_14977_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14977_cast = slice_by_index(begin = var_14977_begin_0, end = var_14977_end_0, end_mask = var_14977_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_14977_cast")];
            tensor<int32, [4]> var_14981_begin_0 = const()[name = tensor<string, []>("op_14981_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_14981_end_0 = const()[name = tensor<string, []>("op_14981_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_14981_end_mask_0 = const()[name = tensor<string, []>("op_14981_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_14981_cast = slice_by_index(begin = var_14981_begin_0, end = var_14981_end_0, end_mask = var_14981_end_mask_0, x = q_61_cast)[name = tensor<string, []>("op_14981_cast")];
            tensor<int32, [4]> var_14984_begin_0 = const()[name = tensor<string, []>("op_14984_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14984_end_0 = const()[name = tensor<string, []>("op_14984_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14984_end_mask_0 = const()[name = tensor<string, []>("op_14984_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14984_cast = slice_by_index(begin = var_14984_begin_0, end = var_14984_end_0, end_mask = var_14984_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14984_cast")];
            tensor<int32, [4]> var_14985_begin_0 = const()[name = tensor<string, []>("op_14985_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14985_end_0 = const()[name = tensor<string, []>("op_14985_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14985_end_mask_0 = const()[name = tensor<string, []>("op_14985_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14985_cast = slice_by_index(begin = var_14985_begin_0, end = var_14985_end_0, end_mask = var_14985_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14985_cast")];
            tensor<int32, [4]> var_14986_begin_0 = const()[name = tensor<string, []>("op_14986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14986_end_0 = const()[name = tensor<string, []>("op_14986_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14986_end_mask_0 = const()[name = tensor<string, []>("op_14986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14986_cast = slice_by_index(begin = var_14986_begin_0, end = var_14986_end_0, end_mask = var_14986_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14986_cast")];
            tensor<int32, [4]> var_14987_begin_0 = const()[name = tensor<string, []>("op_14987_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14987_end_0 = const()[name = tensor<string, []>("op_14987_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14987_end_mask_0 = const()[name = tensor<string, []>("op_14987_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14987_cast = slice_by_index(begin = var_14987_begin_0, end = var_14987_end_0, end_mask = var_14987_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14987_cast")];
            tensor<int32, [4]> var_14988_begin_0 = const()[name = tensor<string, []>("op_14988_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14988_end_0 = const()[name = tensor<string, []>("op_14988_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14988_end_mask_0 = const()[name = tensor<string, []>("op_14988_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14988_cast = slice_by_index(begin = var_14988_begin_0, end = var_14988_end_0, end_mask = var_14988_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14988_cast")];
            tensor<int32, [4]> var_14989_begin_0 = const()[name = tensor<string, []>("op_14989_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14989_end_0 = const()[name = tensor<string, []>("op_14989_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14989_end_mask_0 = const()[name = tensor<string, []>("op_14989_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14989_cast = slice_by_index(begin = var_14989_begin_0, end = var_14989_end_0, end_mask = var_14989_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14989_cast")];
            tensor<int32, [4]> var_14990_begin_0 = const()[name = tensor<string, []>("op_14990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14990_end_0 = const()[name = tensor<string, []>("op_14990_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14990_end_mask_0 = const()[name = tensor<string, []>("op_14990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14990_cast = slice_by_index(begin = var_14990_begin_0, end = var_14990_end_0, end_mask = var_14990_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14990_cast")];
            tensor<int32, [4]> var_14991_begin_0 = const()[name = tensor<string, []>("op_14991_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14991_end_0 = const()[name = tensor<string, []>("op_14991_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14991_end_mask_0 = const()[name = tensor<string, []>("op_14991_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14991_cast = slice_by_index(begin = var_14991_begin_0, end = var_14991_end_0, end_mask = var_14991_end_mask_0, x = var_14965_cast)[name = tensor<string, []>("op_14991_cast")];
            tensor<int32, [4]> var_14992_begin_0 = const()[name = tensor<string, []>("op_14992_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_14992_end_0 = const()[name = tensor<string, []>("op_14992_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_14992_end_mask_0 = const()[name = tensor<string, []>("op_14992_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14992_cast = slice_by_index(begin = var_14992_begin_0, end = var_14992_end_0, end_mask = var_14992_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14992_cast")];
            tensor<int32, [4]> var_14993_begin_0 = const()[name = tensor<string, []>("op_14993_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_14993_end_0 = const()[name = tensor<string, []>("op_14993_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_14993_end_mask_0 = const()[name = tensor<string, []>("op_14993_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14993_cast = slice_by_index(begin = var_14993_begin_0, end = var_14993_end_0, end_mask = var_14993_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14993_cast")];
            tensor<int32, [4]> var_14994_begin_0 = const()[name = tensor<string, []>("op_14994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_14994_end_0 = const()[name = tensor<string, []>("op_14994_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_14994_end_mask_0 = const()[name = tensor<string, []>("op_14994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14994_cast = slice_by_index(begin = var_14994_begin_0, end = var_14994_end_0, end_mask = var_14994_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14994_cast")];
            tensor<int32, [4]> var_14995_begin_0 = const()[name = tensor<string, []>("op_14995_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_14995_end_0 = const()[name = tensor<string, []>("op_14995_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_14995_end_mask_0 = const()[name = tensor<string, []>("op_14995_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14995_cast = slice_by_index(begin = var_14995_begin_0, end = var_14995_end_0, end_mask = var_14995_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14995_cast")];
            tensor<int32, [4]> var_14996_begin_0 = const()[name = tensor<string, []>("op_14996_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_14996_end_0 = const()[name = tensor<string, []>("op_14996_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_14996_end_mask_0 = const()[name = tensor<string, []>("op_14996_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14996_cast = slice_by_index(begin = var_14996_begin_0, end = var_14996_end_0, end_mask = var_14996_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14996_cast")];
            tensor<int32, [4]> var_14997_begin_0 = const()[name = tensor<string, []>("op_14997_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_14997_end_0 = const()[name = tensor<string, []>("op_14997_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_14997_end_mask_0 = const()[name = tensor<string, []>("op_14997_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14997_cast = slice_by_index(begin = var_14997_begin_0, end = var_14997_end_0, end_mask = var_14997_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14997_cast")];
            tensor<int32, [4]> var_14998_begin_0 = const()[name = tensor<string, []>("op_14998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_14998_end_0 = const()[name = tensor<string, []>("op_14998_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_14998_end_mask_0 = const()[name = tensor<string, []>("op_14998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14998_cast = slice_by_index(begin = var_14998_begin_0, end = var_14998_end_0, end_mask = var_14998_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14998_cast")];
            tensor<int32, [4]> var_14999_begin_0 = const()[name = tensor<string, []>("op_14999_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_14999_end_0 = const()[name = tensor<string, []>("op_14999_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_14999_end_mask_0 = const()[name = tensor<string, []>("op_14999_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_14999_cast = slice_by_index(begin = var_14999_begin_0, end = var_14999_end_0, end_mask = var_14999_end_mask_0, x = var_14969_cast)[name = tensor<string, []>("op_14999_cast")];
            tensor<int32, [4]> var_15000_begin_0 = const()[name = tensor<string, []>("op_15000_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15000_end_0 = const()[name = tensor<string, []>("op_15000_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15000_end_mask_0 = const()[name = tensor<string, []>("op_15000_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15000_cast = slice_by_index(begin = var_15000_begin_0, end = var_15000_end_0, end_mask = var_15000_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15000_cast")];
            tensor<int32, [4]> var_15001_begin_0 = const()[name = tensor<string, []>("op_15001_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15001_end_0 = const()[name = tensor<string, []>("op_15001_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15001_end_mask_0 = const()[name = tensor<string, []>("op_15001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15001_cast = slice_by_index(begin = var_15001_begin_0, end = var_15001_end_0, end_mask = var_15001_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15001_cast")];
            tensor<int32, [4]> var_15002_begin_0 = const()[name = tensor<string, []>("op_15002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15002_end_0 = const()[name = tensor<string, []>("op_15002_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15002_end_mask_0 = const()[name = tensor<string, []>("op_15002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15002_cast = slice_by_index(begin = var_15002_begin_0, end = var_15002_end_0, end_mask = var_15002_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15002_cast")];
            tensor<int32, [4]> var_15003_begin_0 = const()[name = tensor<string, []>("op_15003_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15003_end_0 = const()[name = tensor<string, []>("op_15003_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15003_end_mask_0 = const()[name = tensor<string, []>("op_15003_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15003_cast = slice_by_index(begin = var_15003_begin_0, end = var_15003_end_0, end_mask = var_15003_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15003_cast")];
            tensor<int32, [4]> var_15004_begin_0 = const()[name = tensor<string, []>("op_15004_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15004_end_0 = const()[name = tensor<string, []>("op_15004_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15004_end_mask_0 = const()[name = tensor<string, []>("op_15004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15004_cast = slice_by_index(begin = var_15004_begin_0, end = var_15004_end_0, end_mask = var_15004_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15004_cast")];
            tensor<int32, [4]> var_15005_begin_0 = const()[name = tensor<string, []>("op_15005_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15005_end_0 = const()[name = tensor<string, []>("op_15005_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15005_end_mask_0 = const()[name = tensor<string, []>("op_15005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15005_cast = slice_by_index(begin = var_15005_begin_0, end = var_15005_end_0, end_mask = var_15005_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15005_cast")];
            tensor<int32, [4]> var_15006_begin_0 = const()[name = tensor<string, []>("op_15006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15006_end_0 = const()[name = tensor<string, []>("op_15006_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15006_end_mask_0 = const()[name = tensor<string, []>("op_15006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15006_cast = slice_by_index(begin = var_15006_begin_0, end = var_15006_end_0, end_mask = var_15006_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15006_cast")];
            tensor<int32, [4]> var_15007_begin_0 = const()[name = tensor<string, []>("op_15007_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15007_end_0 = const()[name = tensor<string, []>("op_15007_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15007_end_mask_0 = const()[name = tensor<string, []>("op_15007_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15007_cast = slice_by_index(begin = var_15007_begin_0, end = var_15007_end_0, end_mask = var_15007_end_mask_0, x = var_14973_cast)[name = tensor<string, []>("op_15007_cast")];
            tensor<int32, [4]> var_15008_begin_0 = const()[name = tensor<string, []>("op_15008_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15008_end_0 = const()[name = tensor<string, []>("op_15008_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15008_end_mask_0 = const()[name = tensor<string, []>("op_15008_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15008_cast = slice_by_index(begin = var_15008_begin_0, end = var_15008_end_0, end_mask = var_15008_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15008_cast")];
            tensor<int32, [4]> var_15009_begin_0 = const()[name = tensor<string, []>("op_15009_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15009_end_0 = const()[name = tensor<string, []>("op_15009_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15009_end_mask_0 = const()[name = tensor<string, []>("op_15009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15009_cast = slice_by_index(begin = var_15009_begin_0, end = var_15009_end_0, end_mask = var_15009_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15009_cast")];
            tensor<int32, [4]> var_15010_begin_0 = const()[name = tensor<string, []>("op_15010_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15010_end_0 = const()[name = tensor<string, []>("op_15010_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15010_end_mask_0 = const()[name = tensor<string, []>("op_15010_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15010_cast = slice_by_index(begin = var_15010_begin_0, end = var_15010_end_0, end_mask = var_15010_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15010_cast")];
            tensor<int32, [4]> var_15011_begin_0 = const()[name = tensor<string, []>("op_15011_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15011_end_0 = const()[name = tensor<string, []>("op_15011_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15011_end_mask_0 = const()[name = tensor<string, []>("op_15011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15011_cast = slice_by_index(begin = var_15011_begin_0, end = var_15011_end_0, end_mask = var_15011_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15011_cast")];
            tensor<int32, [4]> var_15012_begin_0 = const()[name = tensor<string, []>("op_15012_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15012_end_0 = const()[name = tensor<string, []>("op_15012_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15012_end_mask_0 = const()[name = tensor<string, []>("op_15012_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15012_cast = slice_by_index(begin = var_15012_begin_0, end = var_15012_end_0, end_mask = var_15012_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15012_cast")];
            tensor<int32, [4]> var_15013_begin_0 = const()[name = tensor<string, []>("op_15013_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15013_end_0 = const()[name = tensor<string, []>("op_15013_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15013_end_mask_0 = const()[name = tensor<string, []>("op_15013_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15013_cast = slice_by_index(begin = var_15013_begin_0, end = var_15013_end_0, end_mask = var_15013_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15013_cast")];
            tensor<int32, [4]> var_15014_begin_0 = const()[name = tensor<string, []>("op_15014_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15014_end_0 = const()[name = tensor<string, []>("op_15014_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15014_end_mask_0 = const()[name = tensor<string, []>("op_15014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15014_cast = slice_by_index(begin = var_15014_begin_0, end = var_15014_end_0, end_mask = var_15014_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15014_cast")];
            tensor<int32, [4]> var_15015_begin_0 = const()[name = tensor<string, []>("op_15015_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15015_end_0 = const()[name = tensor<string, []>("op_15015_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15015_end_mask_0 = const()[name = tensor<string, []>("op_15015_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15015_cast = slice_by_index(begin = var_15015_begin_0, end = var_15015_end_0, end_mask = var_15015_end_mask_0, x = var_14977_cast)[name = tensor<string, []>("op_15015_cast")];
            tensor<int32, [4]> var_15016_begin_0 = const()[name = tensor<string, []>("op_15016_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15016_end_0 = const()[name = tensor<string, []>("op_15016_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15016_end_mask_0 = const()[name = tensor<string, []>("op_15016_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15016_cast = slice_by_index(begin = var_15016_begin_0, end = var_15016_end_0, end_mask = var_15016_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15016_cast")];
            tensor<int32, [4]> var_15017_begin_0 = const()[name = tensor<string, []>("op_15017_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15017_end_0 = const()[name = tensor<string, []>("op_15017_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15017_end_mask_0 = const()[name = tensor<string, []>("op_15017_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15017_cast = slice_by_index(begin = var_15017_begin_0, end = var_15017_end_0, end_mask = var_15017_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15017_cast")];
            tensor<int32, [4]> var_15018_begin_0 = const()[name = tensor<string, []>("op_15018_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15018_end_0 = const()[name = tensor<string, []>("op_15018_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15018_end_mask_0 = const()[name = tensor<string, []>("op_15018_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15018_cast = slice_by_index(begin = var_15018_begin_0, end = var_15018_end_0, end_mask = var_15018_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15018_cast")];
            tensor<int32, [4]> var_15019_begin_0 = const()[name = tensor<string, []>("op_15019_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15019_end_0 = const()[name = tensor<string, []>("op_15019_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15019_end_mask_0 = const()[name = tensor<string, []>("op_15019_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15019_cast = slice_by_index(begin = var_15019_begin_0, end = var_15019_end_0, end_mask = var_15019_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15019_cast")];
            tensor<int32, [4]> var_15020_begin_0 = const()[name = tensor<string, []>("op_15020_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15020_end_0 = const()[name = tensor<string, []>("op_15020_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15020_end_mask_0 = const()[name = tensor<string, []>("op_15020_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15020_cast = slice_by_index(begin = var_15020_begin_0, end = var_15020_end_0, end_mask = var_15020_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15020_cast")];
            tensor<int32, [4]> var_15021_begin_0 = const()[name = tensor<string, []>("op_15021_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15021_end_0 = const()[name = tensor<string, []>("op_15021_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15021_end_mask_0 = const()[name = tensor<string, []>("op_15021_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15021_cast = slice_by_index(begin = var_15021_begin_0, end = var_15021_end_0, end_mask = var_15021_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15021_cast")];
            tensor<int32, [4]> var_15022_begin_0 = const()[name = tensor<string, []>("op_15022_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15022_end_0 = const()[name = tensor<string, []>("op_15022_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15022_end_mask_0 = const()[name = tensor<string, []>("op_15022_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15022_cast = slice_by_index(begin = var_15022_begin_0, end = var_15022_end_0, end_mask = var_15022_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15022_cast")];
            tensor<int32, [4]> var_15023_begin_0 = const()[name = tensor<string, []>("op_15023_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15023_end_0 = const()[name = tensor<string, []>("op_15023_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15023_end_mask_0 = const()[name = tensor<string, []>("op_15023_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15023_cast = slice_by_index(begin = var_15023_begin_0, end = var_15023_end_0, end_mask = var_15023_end_mask_0, x = var_14981_cast)[name = tensor<string, []>("op_15023_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_15028_begin_0 = const()[name = tensor<string, []>("op_15028_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15028_end_0 = const()[name = tensor<string, []>("op_15028_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 64])];
            tensor<bool, [4]> var_15028_end_mask_0 = const()[name = tensor<string, []>("op_15028_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, 64]> var_15028_cast = slice_by_index(begin = var_15028_begin_0, end = var_15028_end_0, end_mask = var_15028_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_15028_cast")];
            tensor<int32, [4]> var_15032_begin_0 = const()[name = tensor<string, []>("op_15032_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_15032_end_0 = const()[name = tensor<string, []>("op_15032_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 128])];
            tensor<bool, [4]> var_15032_end_mask_0 = const()[name = tensor<string, []>("op_15032_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_15032_cast = slice_by_index(begin = var_15032_begin_0, end = var_15032_end_0, end_mask = var_15032_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_15032_cast")];
            tensor<int32, [4]> var_15036_begin_0 = const()[name = tensor<string, []>("op_15036_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_15036_end_0 = const()[name = tensor<string, []>("op_15036_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 192])];
            tensor<bool, [4]> var_15036_end_mask_0 = const()[name = tensor<string, []>("op_15036_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_15036_cast = slice_by_index(begin = var_15036_begin_0, end = var_15036_end_0, end_mask = var_15036_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_15036_cast")];
            tensor<int32, [4]> var_15040_begin_0 = const()[name = tensor<string, []>("op_15040_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_15040_end_0 = const()[name = tensor<string, []>("op_15040_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 256])];
            tensor<bool, [4]> var_15040_end_mask_0 = const()[name = tensor<string, []>("op_15040_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_15040_cast = slice_by_index(begin = var_15040_begin_0, end = var_15040_end_0, end_mask = var_15040_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_15040_cast")];
            tensor<int32, [4]> var_15044_begin_0 = const()[name = tensor<string, []>("op_15044_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_15044_end_0 = const()[name = tensor<string, []>("op_15044_end_0"), val = tensor<int32, [4]>([2, 4096, 1, 320])];
            tensor<bool, [4]> var_15044_end_mask_0 = const()[name = tensor<string, []>("op_15044_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 4096, 1, 64]> var_15044_cast = slice_by_index(begin = var_15044_begin_0, end = var_15044_end_0, end_mask = var_15044_end_mask_0, x = transpose_1)[name = tensor<string, []>("op_15044_cast")];
            tensor<int32, [4]> var_15046_begin_0 = const()[name = tensor<string, []>("op_15046_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15046_end_0 = const()[name = tensor<string, []>("op_15046_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15046_end_mask_0 = const()[name = tensor<string, []>("op_15046_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15046_cast = slice_by_index(begin = var_15046_begin_0, end = var_15046_end_0, end_mask = var_15046_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_15046_cast")];
            tensor<int32, [4]> var_15050_begin_0 = const()[name = tensor<string, []>("op_15050_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15050_end_0 = const()[name = tensor<string, []>("op_15050_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_15050_end_mask_0 = const()[name = tensor<string, []>("op_15050_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15050_cast = slice_by_index(begin = var_15050_begin_0, end = var_15050_end_0, end_mask = var_15050_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_15050_cast")];
            tensor<int32, [4]> var_15054_begin_0 = const()[name = tensor<string, []>("op_15054_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15054_end_0 = const()[name = tensor<string, []>("op_15054_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_15054_end_mask_0 = const()[name = tensor<string, []>("op_15054_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15054_cast = slice_by_index(begin = var_15054_begin_0, end = var_15054_end_0, end_mask = var_15054_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_15054_cast")];
            tensor<int32, [4]> var_15058_begin_0 = const()[name = tensor<string, []>("op_15058_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15058_end_0 = const()[name = tensor<string, []>("op_15058_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_15058_end_mask_0 = const()[name = tensor<string, []>("op_15058_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15058_cast = slice_by_index(begin = var_15058_begin_0, end = var_15058_end_0, end_mask = var_15058_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_15058_cast")];
            tensor<int32, [4]> var_15062_begin_0 = const()[name = tensor<string, []>("op_15062_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15062_end_0 = const()[name = tensor<string, []>("op_15062_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_15062_end_mask_0 = const()[name = tensor<string, []>("op_15062_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15062_cast = slice_by_index(begin = var_15062_begin_0, end = var_15062_end_0, end_mask = var_15062_end_mask_0, x = v_61_cast)[name = tensor<string, []>("op_15062_cast")];
            tensor<string, []> var_15066_equation_0 = const()[name = tensor<string, []>("op_15066_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15066_cast = einsum(equation = var_15066_equation_0, values = (var_15028_cast, var_14984_cast))[name = tensor<string, []>("op_15066_cast")];
            tensor<fp16, []> var_15067_to_fp16 = const()[name = tensor<string, []>("op_15067_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1041_cast = mul(x = var_15066_cast, y = var_15067_to_fp16)[name = tensor<string, []>("aw_chunk_1041_cast")];
            tensor<string, []> var_15070_equation_0 = const()[name = tensor<string, []>("op_15070_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15070_cast = einsum(equation = var_15070_equation_0, values = (var_15028_cast, var_14985_cast))[name = tensor<string, []>("op_15070_cast")];
            tensor<fp16, []> var_15071_to_fp16 = const()[name = tensor<string, []>("op_15071_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1043_cast = mul(x = var_15070_cast, y = var_15071_to_fp16)[name = tensor<string, []>("aw_chunk_1043_cast")];
            tensor<string, []> var_15074_equation_0 = const()[name = tensor<string, []>("op_15074_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15074_cast = einsum(equation = var_15074_equation_0, values = (var_15028_cast, var_14986_cast))[name = tensor<string, []>("op_15074_cast")];
            tensor<fp16, []> var_15075_to_fp16 = const()[name = tensor<string, []>("op_15075_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1045_cast = mul(x = var_15074_cast, y = var_15075_to_fp16)[name = tensor<string, []>("aw_chunk_1045_cast")];
            tensor<string, []> var_15078_equation_0 = const()[name = tensor<string, []>("op_15078_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15078_cast = einsum(equation = var_15078_equation_0, values = (var_15028_cast, var_14987_cast))[name = tensor<string, []>("op_15078_cast")];
            tensor<fp16, []> var_15079_to_fp16 = const()[name = tensor<string, []>("op_15079_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1047_cast = mul(x = var_15078_cast, y = var_15079_to_fp16)[name = tensor<string, []>("aw_chunk_1047_cast")];
            tensor<string, []> var_15082_equation_0 = const()[name = tensor<string, []>("op_15082_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15082_cast = einsum(equation = var_15082_equation_0, values = (var_15028_cast, var_14988_cast))[name = tensor<string, []>("op_15082_cast")];
            tensor<fp16, []> var_15083_to_fp16 = const()[name = tensor<string, []>("op_15083_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1049_cast = mul(x = var_15082_cast, y = var_15083_to_fp16)[name = tensor<string, []>("aw_chunk_1049_cast")];
            tensor<string, []> var_15086_equation_0 = const()[name = tensor<string, []>("op_15086_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15086_cast = einsum(equation = var_15086_equation_0, values = (var_15028_cast, var_14989_cast))[name = tensor<string, []>("op_15086_cast")];
            tensor<fp16, []> var_15087_to_fp16 = const()[name = tensor<string, []>("op_15087_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1051_cast = mul(x = var_15086_cast, y = var_15087_to_fp16)[name = tensor<string, []>("aw_chunk_1051_cast")];
            tensor<string, []> var_15090_equation_0 = const()[name = tensor<string, []>("op_15090_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15090_cast = einsum(equation = var_15090_equation_0, values = (var_15028_cast, var_14990_cast))[name = tensor<string, []>("op_15090_cast")];
            tensor<fp16, []> var_15091_to_fp16 = const()[name = tensor<string, []>("op_15091_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1053_cast = mul(x = var_15090_cast, y = var_15091_to_fp16)[name = tensor<string, []>("aw_chunk_1053_cast")];
            tensor<string, []> var_15094_equation_0 = const()[name = tensor<string, []>("op_15094_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15094_cast = einsum(equation = var_15094_equation_0, values = (var_15028_cast, var_14991_cast))[name = tensor<string, []>("op_15094_cast")];
            tensor<fp16, []> var_15095_to_fp16 = const()[name = tensor<string, []>("op_15095_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1055_cast = mul(x = var_15094_cast, y = var_15095_to_fp16)[name = tensor<string, []>("aw_chunk_1055_cast")];
            tensor<string, []> var_15098_equation_0 = const()[name = tensor<string, []>("op_15098_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15098_cast = einsum(equation = var_15098_equation_0, values = (var_15032_cast, var_14992_cast))[name = tensor<string, []>("op_15098_cast")];
            tensor<fp16, []> var_15099_to_fp16 = const()[name = tensor<string, []>("op_15099_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1057_cast = mul(x = var_15098_cast, y = var_15099_to_fp16)[name = tensor<string, []>("aw_chunk_1057_cast")];
            tensor<string, []> var_15102_equation_0 = const()[name = tensor<string, []>("op_15102_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15102_cast = einsum(equation = var_15102_equation_0, values = (var_15032_cast, var_14993_cast))[name = tensor<string, []>("op_15102_cast")];
            tensor<fp16, []> var_15103_to_fp16 = const()[name = tensor<string, []>("op_15103_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1059_cast = mul(x = var_15102_cast, y = var_15103_to_fp16)[name = tensor<string, []>("aw_chunk_1059_cast")];
            tensor<string, []> var_15106_equation_0 = const()[name = tensor<string, []>("op_15106_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15106_cast = einsum(equation = var_15106_equation_0, values = (var_15032_cast, var_14994_cast))[name = tensor<string, []>("op_15106_cast")];
            tensor<fp16, []> var_15107_to_fp16 = const()[name = tensor<string, []>("op_15107_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1061_cast = mul(x = var_15106_cast, y = var_15107_to_fp16)[name = tensor<string, []>("aw_chunk_1061_cast")];
            tensor<string, []> var_15110_equation_0 = const()[name = tensor<string, []>("op_15110_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15110_cast = einsum(equation = var_15110_equation_0, values = (var_15032_cast, var_14995_cast))[name = tensor<string, []>("op_15110_cast")];
            tensor<fp16, []> var_15111_to_fp16 = const()[name = tensor<string, []>("op_15111_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1063_cast = mul(x = var_15110_cast, y = var_15111_to_fp16)[name = tensor<string, []>("aw_chunk_1063_cast")];
            tensor<string, []> var_15114_equation_0 = const()[name = tensor<string, []>("op_15114_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15114_cast = einsum(equation = var_15114_equation_0, values = (var_15032_cast, var_14996_cast))[name = tensor<string, []>("op_15114_cast")];
            tensor<fp16, []> var_15115_to_fp16 = const()[name = tensor<string, []>("op_15115_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1065_cast = mul(x = var_15114_cast, y = var_15115_to_fp16)[name = tensor<string, []>("aw_chunk_1065_cast")];
            tensor<string, []> var_15118_equation_0 = const()[name = tensor<string, []>("op_15118_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15118_cast = einsum(equation = var_15118_equation_0, values = (var_15032_cast, var_14997_cast))[name = tensor<string, []>("op_15118_cast")];
            tensor<fp16, []> var_15119_to_fp16 = const()[name = tensor<string, []>("op_15119_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1067_cast = mul(x = var_15118_cast, y = var_15119_to_fp16)[name = tensor<string, []>("aw_chunk_1067_cast")];
            tensor<string, []> var_15122_equation_0 = const()[name = tensor<string, []>("op_15122_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15122_cast = einsum(equation = var_15122_equation_0, values = (var_15032_cast, var_14998_cast))[name = tensor<string, []>("op_15122_cast")];
            tensor<fp16, []> var_15123_to_fp16 = const()[name = tensor<string, []>("op_15123_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1069_cast = mul(x = var_15122_cast, y = var_15123_to_fp16)[name = tensor<string, []>("aw_chunk_1069_cast")];
            tensor<string, []> var_15126_equation_0 = const()[name = tensor<string, []>("op_15126_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15126_cast = einsum(equation = var_15126_equation_0, values = (var_15032_cast, var_14999_cast))[name = tensor<string, []>("op_15126_cast")];
            tensor<fp16, []> var_15127_to_fp16 = const()[name = tensor<string, []>("op_15127_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1071_cast = mul(x = var_15126_cast, y = var_15127_to_fp16)[name = tensor<string, []>("aw_chunk_1071_cast")];
            tensor<string, []> var_15130_equation_0 = const()[name = tensor<string, []>("op_15130_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15130_cast = einsum(equation = var_15130_equation_0, values = (var_15036_cast, var_15000_cast))[name = tensor<string, []>("op_15130_cast")];
            tensor<fp16, []> var_15131_to_fp16 = const()[name = tensor<string, []>("op_15131_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1073_cast = mul(x = var_15130_cast, y = var_15131_to_fp16)[name = tensor<string, []>("aw_chunk_1073_cast")];
            tensor<string, []> var_15134_equation_0 = const()[name = tensor<string, []>("op_15134_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15134_cast = einsum(equation = var_15134_equation_0, values = (var_15036_cast, var_15001_cast))[name = tensor<string, []>("op_15134_cast")];
            tensor<fp16, []> var_15135_to_fp16 = const()[name = tensor<string, []>("op_15135_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1075_cast = mul(x = var_15134_cast, y = var_15135_to_fp16)[name = tensor<string, []>("aw_chunk_1075_cast")];
            tensor<string, []> var_15138_equation_0 = const()[name = tensor<string, []>("op_15138_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15138_cast = einsum(equation = var_15138_equation_0, values = (var_15036_cast, var_15002_cast))[name = tensor<string, []>("op_15138_cast")];
            tensor<fp16, []> var_15139_to_fp16 = const()[name = tensor<string, []>("op_15139_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1077_cast = mul(x = var_15138_cast, y = var_15139_to_fp16)[name = tensor<string, []>("aw_chunk_1077_cast")];
            tensor<string, []> var_15142_equation_0 = const()[name = tensor<string, []>("op_15142_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15142_cast = einsum(equation = var_15142_equation_0, values = (var_15036_cast, var_15003_cast))[name = tensor<string, []>("op_15142_cast")];
            tensor<fp16, []> var_15143_to_fp16 = const()[name = tensor<string, []>("op_15143_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1079_cast = mul(x = var_15142_cast, y = var_15143_to_fp16)[name = tensor<string, []>("aw_chunk_1079_cast")];
            tensor<string, []> var_15146_equation_0 = const()[name = tensor<string, []>("op_15146_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15146_cast = einsum(equation = var_15146_equation_0, values = (var_15036_cast, var_15004_cast))[name = tensor<string, []>("op_15146_cast")];
            tensor<fp16, []> var_15147_to_fp16 = const()[name = tensor<string, []>("op_15147_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1081_cast = mul(x = var_15146_cast, y = var_15147_to_fp16)[name = tensor<string, []>("aw_chunk_1081_cast")];
            tensor<string, []> var_15150_equation_0 = const()[name = tensor<string, []>("op_15150_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15150_cast = einsum(equation = var_15150_equation_0, values = (var_15036_cast, var_15005_cast))[name = tensor<string, []>("op_15150_cast")];
            tensor<fp16, []> var_15151_to_fp16 = const()[name = tensor<string, []>("op_15151_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1083_cast = mul(x = var_15150_cast, y = var_15151_to_fp16)[name = tensor<string, []>("aw_chunk_1083_cast")];
            tensor<string, []> var_15154_equation_0 = const()[name = tensor<string, []>("op_15154_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15154_cast = einsum(equation = var_15154_equation_0, values = (var_15036_cast, var_15006_cast))[name = tensor<string, []>("op_15154_cast")];
            tensor<fp16, []> var_15155_to_fp16 = const()[name = tensor<string, []>("op_15155_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1085_cast = mul(x = var_15154_cast, y = var_15155_to_fp16)[name = tensor<string, []>("aw_chunk_1085_cast")];
            tensor<string, []> var_15158_equation_0 = const()[name = tensor<string, []>("op_15158_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15158_cast = einsum(equation = var_15158_equation_0, values = (var_15036_cast, var_15007_cast))[name = tensor<string, []>("op_15158_cast")];
            tensor<fp16, []> var_15159_to_fp16 = const()[name = tensor<string, []>("op_15159_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1087_cast = mul(x = var_15158_cast, y = var_15159_to_fp16)[name = tensor<string, []>("aw_chunk_1087_cast")];
            tensor<string, []> var_15162_equation_0 = const()[name = tensor<string, []>("op_15162_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15162_cast = einsum(equation = var_15162_equation_0, values = (var_15040_cast, var_15008_cast))[name = tensor<string, []>("op_15162_cast")];
            tensor<fp16, []> var_15163_to_fp16 = const()[name = tensor<string, []>("op_15163_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1089_cast = mul(x = var_15162_cast, y = var_15163_to_fp16)[name = tensor<string, []>("aw_chunk_1089_cast")];
            tensor<string, []> var_15166_equation_0 = const()[name = tensor<string, []>("op_15166_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15166_cast = einsum(equation = var_15166_equation_0, values = (var_15040_cast, var_15009_cast))[name = tensor<string, []>("op_15166_cast")];
            tensor<fp16, []> var_15167_to_fp16 = const()[name = tensor<string, []>("op_15167_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1091_cast = mul(x = var_15166_cast, y = var_15167_to_fp16)[name = tensor<string, []>("aw_chunk_1091_cast")];
            tensor<string, []> var_15170_equation_0 = const()[name = tensor<string, []>("op_15170_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15170_cast = einsum(equation = var_15170_equation_0, values = (var_15040_cast, var_15010_cast))[name = tensor<string, []>("op_15170_cast")];
            tensor<fp16, []> var_15171_to_fp16 = const()[name = tensor<string, []>("op_15171_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1093_cast = mul(x = var_15170_cast, y = var_15171_to_fp16)[name = tensor<string, []>("aw_chunk_1093_cast")];
            tensor<string, []> var_15174_equation_0 = const()[name = tensor<string, []>("op_15174_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15174_cast = einsum(equation = var_15174_equation_0, values = (var_15040_cast, var_15011_cast))[name = tensor<string, []>("op_15174_cast")];
            tensor<fp16, []> var_15175_to_fp16 = const()[name = tensor<string, []>("op_15175_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1095_cast = mul(x = var_15174_cast, y = var_15175_to_fp16)[name = tensor<string, []>("aw_chunk_1095_cast")];
            tensor<string, []> var_15178_equation_0 = const()[name = tensor<string, []>("op_15178_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15178_cast = einsum(equation = var_15178_equation_0, values = (var_15040_cast, var_15012_cast))[name = tensor<string, []>("op_15178_cast")];
            tensor<fp16, []> var_15179_to_fp16 = const()[name = tensor<string, []>("op_15179_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1097_cast = mul(x = var_15178_cast, y = var_15179_to_fp16)[name = tensor<string, []>("aw_chunk_1097_cast")];
            tensor<string, []> var_15182_equation_0 = const()[name = tensor<string, []>("op_15182_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15182_cast = einsum(equation = var_15182_equation_0, values = (var_15040_cast, var_15013_cast))[name = tensor<string, []>("op_15182_cast")];
            tensor<fp16, []> var_15183_to_fp16 = const()[name = tensor<string, []>("op_15183_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1099_cast = mul(x = var_15182_cast, y = var_15183_to_fp16)[name = tensor<string, []>("aw_chunk_1099_cast")];
            tensor<string, []> var_15186_equation_0 = const()[name = tensor<string, []>("op_15186_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15186_cast = einsum(equation = var_15186_equation_0, values = (var_15040_cast, var_15014_cast))[name = tensor<string, []>("op_15186_cast")];
            tensor<fp16, []> var_15187_to_fp16 = const()[name = tensor<string, []>("op_15187_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1101_cast = mul(x = var_15186_cast, y = var_15187_to_fp16)[name = tensor<string, []>("aw_chunk_1101_cast")];
            tensor<string, []> var_15190_equation_0 = const()[name = tensor<string, []>("op_15190_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15190_cast = einsum(equation = var_15190_equation_0, values = (var_15040_cast, var_15015_cast))[name = tensor<string, []>("op_15190_cast")];
            tensor<fp16, []> var_15191_to_fp16 = const()[name = tensor<string, []>("op_15191_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1103_cast = mul(x = var_15190_cast, y = var_15191_to_fp16)[name = tensor<string, []>("aw_chunk_1103_cast")];
            tensor<string, []> var_15194_equation_0 = const()[name = tensor<string, []>("op_15194_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15194_cast = einsum(equation = var_15194_equation_0, values = (var_15044_cast, var_15016_cast))[name = tensor<string, []>("op_15194_cast")];
            tensor<fp16, []> var_15195_to_fp16 = const()[name = tensor<string, []>("op_15195_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1105_cast = mul(x = var_15194_cast, y = var_15195_to_fp16)[name = tensor<string, []>("aw_chunk_1105_cast")];
            tensor<string, []> var_15198_equation_0 = const()[name = tensor<string, []>("op_15198_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15198_cast = einsum(equation = var_15198_equation_0, values = (var_15044_cast, var_15017_cast))[name = tensor<string, []>("op_15198_cast")];
            tensor<fp16, []> var_15199_to_fp16 = const()[name = tensor<string, []>("op_15199_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1107_cast = mul(x = var_15198_cast, y = var_15199_to_fp16)[name = tensor<string, []>("aw_chunk_1107_cast")];
            tensor<string, []> var_15202_equation_0 = const()[name = tensor<string, []>("op_15202_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15202_cast = einsum(equation = var_15202_equation_0, values = (var_15044_cast, var_15018_cast))[name = tensor<string, []>("op_15202_cast")];
            tensor<fp16, []> var_15203_to_fp16 = const()[name = tensor<string, []>("op_15203_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1109_cast = mul(x = var_15202_cast, y = var_15203_to_fp16)[name = tensor<string, []>("aw_chunk_1109_cast")];
            tensor<string, []> var_15206_equation_0 = const()[name = tensor<string, []>("op_15206_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15206_cast = einsum(equation = var_15206_equation_0, values = (var_15044_cast, var_15019_cast))[name = tensor<string, []>("op_15206_cast")];
            tensor<fp16, []> var_15207_to_fp16 = const()[name = tensor<string, []>("op_15207_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1111_cast = mul(x = var_15206_cast, y = var_15207_to_fp16)[name = tensor<string, []>("aw_chunk_1111_cast")];
            tensor<string, []> var_15210_equation_0 = const()[name = tensor<string, []>("op_15210_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15210_cast = einsum(equation = var_15210_equation_0, values = (var_15044_cast, var_15020_cast))[name = tensor<string, []>("op_15210_cast")];
            tensor<fp16, []> var_15211_to_fp16 = const()[name = tensor<string, []>("op_15211_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1113_cast = mul(x = var_15210_cast, y = var_15211_to_fp16)[name = tensor<string, []>("aw_chunk_1113_cast")];
            tensor<string, []> var_15214_equation_0 = const()[name = tensor<string, []>("op_15214_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15214_cast = einsum(equation = var_15214_equation_0, values = (var_15044_cast, var_15021_cast))[name = tensor<string, []>("op_15214_cast")];
            tensor<fp16, []> var_15215_to_fp16 = const()[name = tensor<string, []>("op_15215_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1115_cast = mul(x = var_15214_cast, y = var_15215_to_fp16)[name = tensor<string, []>("aw_chunk_1115_cast")];
            tensor<string, []> var_15218_equation_0 = const()[name = tensor<string, []>("op_15218_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15218_cast = einsum(equation = var_15218_equation_0, values = (var_15044_cast, var_15022_cast))[name = tensor<string, []>("op_15218_cast")];
            tensor<fp16, []> var_15219_to_fp16 = const()[name = tensor<string, []>("op_15219_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1117_cast = mul(x = var_15218_cast, y = var_15219_to_fp16)[name = tensor<string, []>("aw_chunk_1117_cast")];
            tensor<string, []> var_15222_equation_0 = const()[name = tensor<string, []>("op_15222_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 4096, 1, 512]> var_15222_cast = einsum(equation = var_15222_equation_0, values = (var_15044_cast, var_15023_cast))[name = tensor<string, []>("op_15222_cast")];
            tensor<fp16, []> var_15223_to_fp16 = const()[name = tensor<string, []>("op_15223_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 4096, 1, 512]> aw_chunk_1119_cast = mul(x = var_15222_cast, y = var_15223_to_fp16)[name = tensor<string, []>("aw_chunk_1119_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15225_cast = softmax(axis = var_12807, x = aw_chunk_1041_cast)[name = tensor<string, []>("op_15225_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15226_cast = softmax(axis = var_12807, x = aw_chunk_1043_cast)[name = tensor<string, []>("op_15226_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15227_cast = softmax(axis = var_12807, x = aw_chunk_1045_cast)[name = tensor<string, []>("op_15227_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15228_cast = softmax(axis = var_12807, x = aw_chunk_1047_cast)[name = tensor<string, []>("op_15228_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15229_cast = softmax(axis = var_12807, x = aw_chunk_1049_cast)[name = tensor<string, []>("op_15229_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15230_cast = softmax(axis = var_12807, x = aw_chunk_1051_cast)[name = tensor<string, []>("op_15230_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15231_cast = softmax(axis = var_12807, x = aw_chunk_1053_cast)[name = tensor<string, []>("op_15231_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15232_cast = softmax(axis = var_12807, x = aw_chunk_1055_cast)[name = tensor<string, []>("op_15232_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15233_cast = softmax(axis = var_12807, x = aw_chunk_1057_cast)[name = tensor<string, []>("op_15233_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15234_cast = softmax(axis = var_12807, x = aw_chunk_1059_cast)[name = tensor<string, []>("op_15234_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15235_cast = softmax(axis = var_12807, x = aw_chunk_1061_cast)[name = tensor<string, []>("op_15235_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15236_cast = softmax(axis = var_12807, x = aw_chunk_1063_cast)[name = tensor<string, []>("op_15236_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15237_cast = softmax(axis = var_12807, x = aw_chunk_1065_cast)[name = tensor<string, []>("op_15237_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15238_cast = softmax(axis = var_12807, x = aw_chunk_1067_cast)[name = tensor<string, []>("op_15238_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15239_cast = softmax(axis = var_12807, x = aw_chunk_1069_cast)[name = tensor<string, []>("op_15239_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15240_cast = softmax(axis = var_12807, x = aw_chunk_1071_cast)[name = tensor<string, []>("op_15240_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15241_cast = softmax(axis = var_12807, x = aw_chunk_1073_cast)[name = tensor<string, []>("op_15241_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15242_cast = softmax(axis = var_12807, x = aw_chunk_1075_cast)[name = tensor<string, []>("op_15242_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15243_cast = softmax(axis = var_12807, x = aw_chunk_1077_cast)[name = tensor<string, []>("op_15243_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15244_cast = softmax(axis = var_12807, x = aw_chunk_1079_cast)[name = tensor<string, []>("op_15244_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15245_cast = softmax(axis = var_12807, x = aw_chunk_1081_cast)[name = tensor<string, []>("op_15245_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15246_cast = softmax(axis = var_12807, x = aw_chunk_1083_cast)[name = tensor<string, []>("op_15246_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15247_cast = softmax(axis = var_12807, x = aw_chunk_1085_cast)[name = tensor<string, []>("op_15247_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15248_cast = softmax(axis = var_12807, x = aw_chunk_1087_cast)[name = tensor<string, []>("op_15248_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15249_cast = softmax(axis = var_12807, x = aw_chunk_1089_cast)[name = tensor<string, []>("op_15249_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15250_cast = softmax(axis = var_12807, x = aw_chunk_1091_cast)[name = tensor<string, []>("op_15250_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15251_cast = softmax(axis = var_12807, x = aw_chunk_1093_cast)[name = tensor<string, []>("op_15251_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15252_cast = softmax(axis = var_12807, x = aw_chunk_1095_cast)[name = tensor<string, []>("op_15252_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15253_cast = softmax(axis = var_12807, x = aw_chunk_1097_cast)[name = tensor<string, []>("op_15253_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15254_cast = softmax(axis = var_12807, x = aw_chunk_1099_cast)[name = tensor<string, []>("op_15254_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15255_cast = softmax(axis = var_12807, x = aw_chunk_1101_cast)[name = tensor<string, []>("op_15255_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15256_cast = softmax(axis = var_12807, x = aw_chunk_1103_cast)[name = tensor<string, []>("op_15256_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15257_cast = softmax(axis = var_12807, x = aw_chunk_1105_cast)[name = tensor<string, []>("op_15257_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15258_cast = softmax(axis = var_12807, x = aw_chunk_1107_cast)[name = tensor<string, []>("op_15258_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15259_cast = softmax(axis = var_12807, x = aw_chunk_1109_cast)[name = tensor<string, []>("op_15259_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15260_cast = softmax(axis = var_12807, x = aw_chunk_1111_cast)[name = tensor<string, []>("op_15260_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15261_cast = softmax(axis = var_12807, x = aw_chunk_1113_cast)[name = tensor<string, []>("op_15261_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15262_cast = softmax(axis = var_12807, x = aw_chunk_1115_cast)[name = tensor<string, []>("op_15262_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15263_cast = softmax(axis = var_12807, x = aw_chunk_1117_cast)[name = tensor<string, []>("op_15263_cast")];
            tensor<fp16, [2, 4096, 1, 512]> var_15264_cast = softmax(axis = var_12807, x = aw_chunk_1119_cast)[name = tensor<string, []>("op_15264_cast")];
            tensor<string, []> var_15266_equation_0 = const()[name = tensor<string, []>("op_15266_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15266_cast = einsum(equation = var_15266_equation_0, values = (var_15046_cast, var_15225_cast))[name = tensor<string, []>("op_15266_cast")];
            tensor<string, []> var_15268_equation_0 = const()[name = tensor<string, []>("op_15268_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15268_cast = einsum(equation = var_15268_equation_0, values = (var_15046_cast, var_15226_cast))[name = tensor<string, []>("op_15268_cast")];
            tensor<string, []> var_15270_equation_0 = const()[name = tensor<string, []>("op_15270_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15270_cast = einsum(equation = var_15270_equation_0, values = (var_15046_cast, var_15227_cast))[name = tensor<string, []>("op_15270_cast")];
            tensor<string, []> var_15272_equation_0 = const()[name = tensor<string, []>("op_15272_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15272_cast = einsum(equation = var_15272_equation_0, values = (var_15046_cast, var_15228_cast))[name = tensor<string, []>("op_15272_cast")];
            tensor<string, []> var_15274_equation_0 = const()[name = tensor<string, []>("op_15274_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15274_cast = einsum(equation = var_15274_equation_0, values = (var_15046_cast, var_15229_cast))[name = tensor<string, []>("op_15274_cast")];
            tensor<string, []> var_15276_equation_0 = const()[name = tensor<string, []>("op_15276_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15276_cast = einsum(equation = var_15276_equation_0, values = (var_15046_cast, var_15230_cast))[name = tensor<string, []>("op_15276_cast")];
            tensor<string, []> var_15278_equation_0 = const()[name = tensor<string, []>("op_15278_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15278_cast = einsum(equation = var_15278_equation_0, values = (var_15046_cast, var_15231_cast))[name = tensor<string, []>("op_15278_cast")];
            tensor<string, []> var_15280_equation_0 = const()[name = tensor<string, []>("op_15280_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15280_cast = einsum(equation = var_15280_equation_0, values = (var_15046_cast, var_15232_cast))[name = tensor<string, []>("op_15280_cast")];
            tensor<string, []> var_15282_equation_0 = const()[name = tensor<string, []>("op_15282_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15282_cast = einsum(equation = var_15282_equation_0, values = (var_15050_cast, var_15233_cast))[name = tensor<string, []>("op_15282_cast")];
            tensor<string, []> var_15284_equation_0 = const()[name = tensor<string, []>("op_15284_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15284_cast = einsum(equation = var_15284_equation_0, values = (var_15050_cast, var_15234_cast))[name = tensor<string, []>("op_15284_cast")];
            tensor<string, []> var_15286_equation_0 = const()[name = tensor<string, []>("op_15286_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15286_cast = einsum(equation = var_15286_equation_0, values = (var_15050_cast, var_15235_cast))[name = tensor<string, []>("op_15286_cast")];
            tensor<string, []> var_15288_equation_0 = const()[name = tensor<string, []>("op_15288_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15288_cast = einsum(equation = var_15288_equation_0, values = (var_15050_cast, var_15236_cast))[name = tensor<string, []>("op_15288_cast")];
            tensor<string, []> var_15290_equation_0 = const()[name = tensor<string, []>("op_15290_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15290_cast = einsum(equation = var_15290_equation_0, values = (var_15050_cast, var_15237_cast))[name = tensor<string, []>("op_15290_cast")];
            tensor<string, []> var_15292_equation_0 = const()[name = tensor<string, []>("op_15292_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15292_cast = einsum(equation = var_15292_equation_0, values = (var_15050_cast, var_15238_cast))[name = tensor<string, []>("op_15292_cast")];
            tensor<string, []> var_15294_equation_0 = const()[name = tensor<string, []>("op_15294_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15294_cast = einsum(equation = var_15294_equation_0, values = (var_15050_cast, var_15239_cast))[name = tensor<string, []>("op_15294_cast")];
            tensor<string, []> var_15296_equation_0 = const()[name = tensor<string, []>("op_15296_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15296_cast = einsum(equation = var_15296_equation_0, values = (var_15050_cast, var_15240_cast))[name = tensor<string, []>("op_15296_cast")];
            tensor<string, []> var_15298_equation_0 = const()[name = tensor<string, []>("op_15298_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15298_cast = einsum(equation = var_15298_equation_0, values = (var_15054_cast, var_15241_cast))[name = tensor<string, []>("op_15298_cast")];
            tensor<string, []> var_15300_equation_0 = const()[name = tensor<string, []>("op_15300_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15300_cast = einsum(equation = var_15300_equation_0, values = (var_15054_cast, var_15242_cast))[name = tensor<string, []>("op_15300_cast")];
            tensor<string, []> var_15302_equation_0 = const()[name = tensor<string, []>("op_15302_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15302_cast = einsum(equation = var_15302_equation_0, values = (var_15054_cast, var_15243_cast))[name = tensor<string, []>("op_15302_cast")];
            tensor<string, []> var_15304_equation_0 = const()[name = tensor<string, []>("op_15304_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15304_cast = einsum(equation = var_15304_equation_0, values = (var_15054_cast, var_15244_cast))[name = tensor<string, []>("op_15304_cast")];
            tensor<string, []> var_15306_equation_0 = const()[name = tensor<string, []>("op_15306_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15306_cast = einsum(equation = var_15306_equation_0, values = (var_15054_cast, var_15245_cast))[name = tensor<string, []>("op_15306_cast")];
            tensor<string, []> var_15308_equation_0 = const()[name = tensor<string, []>("op_15308_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15308_cast = einsum(equation = var_15308_equation_0, values = (var_15054_cast, var_15246_cast))[name = tensor<string, []>("op_15308_cast")];
            tensor<string, []> var_15310_equation_0 = const()[name = tensor<string, []>("op_15310_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15310_cast = einsum(equation = var_15310_equation_0, values = (var_15054_cast, var_15247_cast))[name = tensor<string, []>("op_15310_cast")];
            tensor<string, []> var_15312_equation_0 = const()[name = tensor<string, []>("op_15312_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15312_cast = einsum(equation = var_15312_equation_0, values = (var_15054_cast, var_15248_cast))[name = tensor<string, []>("op_15312_cast")];
            tensor<string, []> var_15314_equation_0 = const()[name = tensor<string, []>("op_15314_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15314_cast = einsum(equation = var_15314_equation_0, values = (var_15058_cast, var_15249_cast))[name = tensor<string, []>("op_15314_cast")];
            tensor<string, []> var_15316_equation_0 = const()[name = tensor<string, []>("op_15316_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15316_cast = einsum(equation = var_15316_equation_0, values = (var_15058_cast, var_15250_cast))[name = tensor<string, []>("op_15316_cast")];
            tensor<string, []> var_15318_equation_0 = const()[name = tensor<string, []>("op_15318_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15318_cast = einsum(equation = var_15318_equation_0, values = (var_15058_cast, var_15251_cast))[name = tensor<string, []>("op_15318_cast")];
            tensor<string, []> var_15320_equation_0 = const()[name = tensor<string, []>("op_15320_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15320_cast = einsum(equation = var_15320_equation_0, values = (var_15058_cast, var_15252_cast))[name = tensor<string, []>("op_15320_cast")];
            tensor<string, []> var_15322_equation_0 = const()[name = tensor<string, []>("op_15322_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15322_cast = einsum(equation = var_15322_equation_0, values = (var_15058_cast, var_15253_cast))[name = tensor<string, []>("op_15322_cast")];
            tensor<string, []> var_15324_equation_0 = const()[name = tensor<string, []>("op_15324_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15324_cast = einsum(equation = var_15324_equation_0, values = (var_15058_cast, var_15254_cast))[name = tensor<string, []>("op_15324_cast")];
            tensor<string, []> var_15326_equation_0 = const()[name = tensor<string, []>("op_15326_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15326_cast = einsum(equation = var_15326_equation_0, values = (var_15058_cast, var_15255_cast))[name = tensor<string, []>("op_15326_cast")];
            tensor<string, []> var_15328_equation_0 = const()[name = tensor<string, []>("op_15328_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15328_cast = einsum(equation = var_15328_equation_0, values = (var_15058_cast, var_15256_cast))[name = tensor<string, []>("op_15328_cast")];
            tensor<string, []> var_15330_equation_0 = const()[name = tensor<string, []>("op_15330_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15330_cast = einsum(equation = var_15330_equation_0, values = (var_15062_cast, var_15257_cast))[name = tensor<string, []>("op_15330_cast")];
            tensor<string, []> var_15332_equation_0 = const()[name = tensor<string, []>("op_15332_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15332_cast = einsum(equation = var_15332_equation_0, values = (var_15062_cast, var_15258_cast))[name = tensor<string, []>("op_15332_cast")];
            tensor<string, []> var_15334_equation_0 = const()[name = tensor<string, []>("op_15334_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15334_cast = einsum(equation = var_15334_equation_0, values = (var_15062_cast, var_15259_cast))[name = tensor<string, []>("op_15334_cast")];
            tensor<string, []> var_15336_equation_0 = const()[name = tensor<string, []>("op_15336_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15336_cast = einsum(equation = var_15336_equation_0, values = (var_15062_cast, var_15260_cast))[name = tensor<string, []>("op_15336_cast")];
            tensor<string, []> var_15338_equation_0 = const()[name = tensor<string, []>("op_15338_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15338_cast = einsum(equation = var_15338_equation_0, values = (var_15062_cast, var_15261_cast))[name = tensor<string, []>("op_15338_cast")];
            tensor<string, []> var_15340_equation_0 = const()[name = tensor<string, []>("op_15340_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15340_cast = einsum(equation = var_15340_equation_0, values = (var_15062_cast, var_15262_cast))[name = tensor<string, []>("op_15340_cast")];
            tensor<string, []> var_15342_equation_0 = const()[name = tensor<string, []>("op_15342_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15342_cast = einsum(equation = var_15342_equation_0, values = (var_15062_cast, var_15263_cast))[name = tensor<string, []>("op_15342_cast")];
            tensor<string, []> var_15344_equation_0 = const()[name = tensor<string, []>("op_15344_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15344_cast = einsum(equation = var_15344_equation_0, values = (var_15062_cast, var_15264_cast))[name = tensor<string, []>("op_15344_cast")];
            tensor<bool, []> var_15346_interleave_0 = const()[name = tensor<string, []>("op_15346_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15346_cast = concat(axis = var_12782, interleave = var_15346_interleave_0, values = (var_15266_cast, var_15268_cast, var_15270_cast, var_15272_cast, var_15274_cast, var_15276_cast, var_15278_cast, var_15280_cast))[name = tensor<string, []>("op_15346_cast")];
            tensor<bool, []> var_15348_interleave_0 = const()[name = tensor<string, []>("op_15348_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15348_cast = concat(axis = var_12782, interleave = var_15348_interleave_0, values = (var_15282_cast, var_15284_cast, var_15286_cast, var_15288_cast, var_15290_cast, var_15292_cast, var_15294_cast, var_15296_cast))[name = tensor<string, []>("op_15348_cast")];
            tensor<bool, []> var_15350_interleave_0 = const()[name = tensor<string, []>("op_15350_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15350_cast = concat(axis = var_12782, interleave = var_15350_interleave_0, values = (var_15298_cast, var_15300_cast, var_15302_cast, var_15304_cast, var_15306_cast, var_15308_cast, var_15310_cast, var_15312_cast))[name = tensor<string, []>("op_15350_cast")];
            tensor<bool, []> var_15352_interleave_0 = const()[name = tensor<string, []>("op_15352_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15352_cast = concat(axis = var_12782, interleave = var_15352_interleave_0, values = (var_15314_cast, var_15316_cast, var_15318_cast, var_15320_cast, var_15322_cast, var_15324_cast, var_15326_cast, var_15328_cast))[name = tensor<string, []>("op_15352_cast")];
            tensor<bool, []> var_15354_interleave_0 = const()[name = tensor<string, []>("op_15354_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15354_cast = concat(axis = var_12782, interleave = var_15354_interleave_0, values = (var_15330_cast, var_15332_cast, var_15334_cast, var_15336_cast, var_15338_cast, var_15340_cast, var_15342_cast, var_15344_cast))[name = tensor<string, []>("op_15354_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_12807, interleave = input_517_interleave_0, values = (var_15346_cast, var_15348_cast, var_15350_cast, var_15352_cast, var_15354_cast))[name = tensor<string, []>("input_517_cast")];
            tensor<int32, [2]> var_15360 = const()[name = tensor<string, []>("op_15360"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15362 = const()[name = tensor<string, []>("op_15362"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15364_pad_type_0 = const()[name = tensor<string, []>("op_15364_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15364_pad_0 = const()[name = tensor<string, []>("op_15364_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216517056))), lut = tensor<fp16, [4]>([-0x1.608p-4, -0x1.914p-6, 0x1.92cp-6, 0x1.634p-4]), 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, []>(216542720)))];
            tensor<fp16, [2, 320, 1, 4096]> var_15364_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn1_to_out_0_bias_to_fp16, dilations = var_15362, groups = var_12807, pad = var_15364_pad_0, pad_type = var_15364_pad_type_0, strides = var_15360, 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_15364_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_93_cast = add(x = var_15364_cast, y = inputs_91_cast)[name = tensor<string, []>("inputs_93_cast")];
            tensor<int32, [1]> var_15368 = const()[name = tensor<string, []>("op_15368"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_93_cast = reduce_mean(axes = var_15368, keep_dims = var_12802, 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_15372 = const()[name = tensor<string, []>("op_15372"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_15373_cast = reduce_mean(axes = var_15372, keep_dims = var_12802, x = zero_mean_sq_93_cast)[name = tensor<string, []>("op_15373_cast")];
            tensor<fp16, []> var_15374_to_fp16 = const()[name = tensor<string, []>("op_15374_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_15375_cast = add(x = var_15373_cast, y = var_15374_to_fp16)[name = tensor<string, []>("op_15375_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_15375_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_15379_to_fp16 = const()[name = tensor<string, []>("op_15379_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216543424)))];
            tensor<fp16, [2, 320, 1, 4096]> var_15380_cast = add(x = out_93_cast, y = var_15379_to_fp16)[name = tensor<string, []>("op_15380_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_15382_to_fp16 = const()[name = tensor<string, []>("op_15382_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216544128)))];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_329_cast = mul(x = var_15380_cast, y = var_15382_to_fp16)[name = tensor<string, []>("hidden_states_329_cast")];
            tensor<int32, [2]> var_15389 = const()[name = tensor<string, []>("op_15389"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15391 = const()[name = tensor<string, []>("op_15391"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216544832))), lut = tensor<fp16, [4]>([-0x1.a1cp-4, -0x1.ebcp-6, 0x1.f08p-6, 0x1.a54p-4]), 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_15391, groups = var_12807, pad = q_pad_0, pad_type = q_pad_type_0, strides = var_15389, 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_15395 = const()[name = tensor<string, []>("op_15395"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15397 = const()[name = tensor<string, []>("op_15397"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> k_125_pad_type_0 = const()[name = tensor<string, []>("k_125_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> k_125_pad_0 = const()[name = tensor<string, []>("k_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216570496))), lut = tensor<fp16, [4]>([-0x1.444p-4, -0x1.744p-6, 0x1.5c8p-6, 0x1.3f4p-4]), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_k_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> k_125_cast = conv(dilations = var_15397, groups = var_12807, pad = k_125_pad_0, pad_type = k_125_pad_type_0, strides = var_15395, 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_15401 = const()[name = tensor<string, []>("op_15401"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15403 = const()[name = tensor<string, []>("op_15403"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> v_pad_type_0 = const()[name = tensor<string, []>("v_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> v_pad_0 = const()[name = tensor<string, []>("v_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<fp16, [320, 1024, 1, 1]> up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [81920]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216652480))), lut = tensor<fp16, [4]>([-0x1.844p-6, -0x1.7dp-8, 0x1.9c4p-8, 0x1.93p-6]), name = tensor<string, []>("up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_v_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([320, 1024, 1, 1])];
            tensor<fp16, [2, 320, 1, 77]> v_cast = conv(dilations = var_15403, groups = var_12807, pad = v_pad_0, pad_type = v_pad_type_0, strides = var_15401, 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_15407_begin_0 = const()[name = tensor<string, []>("op_15407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15407_end_0 = const()[name = tensor<string, []>("op_15407_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15407_end_mask_0 = const()[name = tensor<string, []>("op_15407_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15407_cast = slice_by_index(begin = var_15407_begin_0, end = var_15407_end_0, end_mask = var_15407_end_mask_0, x = q_cast)[name = tensor<string, []>("op_15407_cast")];
            tensor<int32, [4]> var_15411_begin_0 = const()[name = tensor<string, []>("op_15411_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15411_end_0 = const()[name = tensor<string, []>("op_15411_end_0"), val = tensor<int32, [4]>([2, 128, 1, 4096])];
            tensor<bool, [4]> var_15411_end_mask_0 = const()[name = tensor<string, []>("op_15411_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15411_cast = slice_by_index(begin = var_15411_begin_0, end = var_15411_end_0, end_mask = var_15411_end_mask_0, x = q_cast)[name = tensor<string, []>("op_15411_cast")];
            tensor<int32, [4]> var_15415_begin_0 = const()[name = tensor<string, []>("op_15415_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15415_end_0 = const()[name = tensor<string, []>("op_15415_end_0"), val = tensor<int32, [4]>([2, 192, 1, 4096])];
            tensor<bool, [4]> var_15415_end_mask_0 = const()[name = tensor<string, []>("op_15415_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15415_cast = slice_by_index(begin = var_15415_begin_0, end = var_15415_end_0, end_mask = var_15415_end_mask_0, x = q_cast)[name = tensor<string, []>("op_15415_cast")];
            tensor<int32, [4]> var_15419_begin_0 = const()[name = tensor<string, []>("op_15419_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15419_end_0 = const()[name = tensor<string, []>("op_15419_end_0"), val = tensor<int32, [4]>([2, 256, 1, 4096])];
            tensor<bool, [4]> var_15419_end_mask_0 = const()[name = tensor<string, []>("op_15419_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15419_cast = slice_by_index(begin = var_15419_begin_0, end = var_15419_end_0, end_mask = var_15419_end_mask_0, x = q_cast)[name = tensor<string, []>("op_15419_cast")];
            tensor<int32, [4]> var_15423_begin_0 = const()[name = tensor<string, []>("op_15423_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15423_end_0 = const()[name = tensor<string, []>("op_15423_end_0"), val = tensor<int32, [4]>([2, 320, 1, 4096])];
            tensor<bool, [4]> var_15423_end_mask_0 = const()[name = tensor<string, []>("op_15423_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 4096]> var_15423_cast = slice_by_index(begin = var_15423_begin_0, end = var_15423_end_0, end_mask = var_15423_end_mask_0, x = q_cast)[name = tensor<string, []>("op_15423_cast")];
            tensor<int32, [4]> var_15426_begin_0 = const()[name = tensor<string, []>("op_15426_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15426_end_0 = const()[name = tensor<string, []>("op_15426_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15426_end_mask_0 = const()[name = tensor<string, []>("op_15426_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15426_cast = slice_by_index(begin = var_15426_begin_0, end = var_15426_end_0, end_mask = var_15426_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15426_cast")];
            tensor<int32, [4]> var_15427_begin_0 = const()[name = tensor<string, []>("op_15427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15427_end_0 = const()[name = tensor<string, []>("op_15427_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15427_end_mask_0 = const()[name = tensor<string, []>("op_15427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15427_cast = slice_by_index(begin = var_15427_begin_0, end = var_15427_end_0, end_mask = var_15427_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15427_cast")];
            tensor<int32, [4]> var_15428_begin_0 = const()[name = tensor<string, []>("op_15428_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15428_end_0 = const()[name = tensor<string, []>("op_15428_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15428_end_mask_0 = const()[name = tensor<string, []>("op_15428_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15428_cast = slice_by_index(begin = var_15428_begin_0, end = var_15428_end_0, end_mask = var_15428_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15428_cast")];
            tensor<int32, [4]> var_15429_begin_0 = const()[name = tensor<string, []>("op_15429_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15429_end_0 = const()[name = tensor<string, []>("op_15429_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15429_end_mask_0 = const()[name = tensor<string, []>("op_15429_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15429_cast = slice_by_index(begin = var_15429_begin_0, end = var_15429_end_0, end_mask = var_15429_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15429_cast")];
            tensor<int32, [4]> var_15430_begin_0 = const()[name = tensor<string, []>("op_15430_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15430_end_0 = const()[name = tensor<string, []>("op_15430_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15430_end_mask_0 = const()[name = tensor<string, []>("op_15430_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15430_cast = slice_by_index(begin = var_15430_begin_0, end = var_15430_end_0, end_mask = var_15430_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15430_cast")];
            tensor<int32, [4]> var_15431_begin_0 = const()[name = tensor<string, []>("op_15431_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15431_end_0 = const()[name = tensor<string, []>("op_15431_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15431_end_mask_0 = const()[name = tensor<string, []>("op_15431_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15431_cast = slice_by_index(begin = var_15431_begin_0, end = var_15431_end_0, end_mask = var_15431_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15431_cast")];
            tensor<int32, [4]> var_15432_begin_0 = const()[name = tensor<string, []>("op_15432_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15432_end_0 = const()[name = tensor<string, []>("op_15432_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15432_end_mask_0 = const()[name = tensor<string, []>("op_15432_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15432_cast = slice_by_index(begin = var_15432_begin_0, end = var_15432_end_0, end_mask = var_15432_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15432_cast")];
            tensor<int32, [4]> var_15433_begin_0 = const()[name = tensor<string, []>("op_15433_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15433_end_0 = const()[name = tensor<string, []>("op_15433_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15433_end_mask_0 = const()[name = tensor<string, []>("op_15433_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15433_cast = slice_by_index(begin = var_15433_begin_0, end = var_15433_end_0, end_mask = var_15433_end_mask_0, x = var_15407_cast)[name = tensor<string, []>("op_15433_cast")];
            tensor<int32, [4]> var_15434_begin_0 = const()[name = tensor<string, []>("op_15434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15434_end_0 = const()[name = tensor<string, []>("op_15434_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15434_end_mask_0 = const()[name = tensor<string, []>("op_15434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15434_cast = slice_by_index(begin = var_15434_begin_0, end = var_15434_end_0, end_mask = var_15434_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15434_cast")];
            tensor<int32, [4]> var_15435_begin_0 = const()[name = tensor<string, []>("op_15435_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15435_end_0 = const()[name = tensor<string, []>("op_15435_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15435_end_mask_0 = const()[name = tensor<string, []>("op_15435_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15435_cast = slice_by_index(begin = var_15435_begin_0, end = var_15435_end_0, end_mask = var_15435_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15435_cast")];
            tensor<int32, [4]> var_15436_begin_0 = const()[name = tensor<string, []>("op_15436_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15436_end_0 = const()[name = tensor<string, []>("op_15436_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15436_end_mask_0 = const()[name = tensor<string, []>("op_15436_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15436_cast = slice_by_index(begin = var_15436_begin_0, end = var_15436_end_0, end_mask = var_15436_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15436_cast")];
            tensor<int32, [4]> var_15437_begin_0 = const()[name = tensor<string, []>("op_15437_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15437_end_0 = const()[name = tensor<string, []>("op_15437_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15437_end_mask_0 = const()[name = tensor<string, []>("op_15437_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15437_cast = slice_by_index(begin = var_15437_begin_0, end = var_15437_end_0, end_mask = var_15437_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15437_cast")];
            tensor<int32, [4]> var_15438_begin_0 = const()[name = tensor<string, []>("op_15438_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15438_end_0 = const()[name = tensor<string, []>("op_15438_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15438_end_mask_0 = const()[name = tensor<string, []>("op_15438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15438_cast = slice_by_index(begin = var_15438_begin_0, end = var_15438_end_0, end_mask = var_15438_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15438_cast")];
            tensor<int32, [4]> var_15439_begin_0 = const()[name = tensor<string, []>("op_15439_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15439_end_0 = const()[name = tensor<string, []>("op_15439_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15439_end_mask_0 = const()[name = tensor<string, []>("op_15439_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15439_cast = slice_by_index(begin = var_15439_begin_0, end = var_15439_end_0, end_mask = var_15439_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15439_cast")];
            tensor<int32, [4]> var_15440_begin_0 = const()[name = tensor<string, []>("op_15440_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15440_end_0 = const()[name = tensor<string, []>("op_15440_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15440_end_mask_0 = const()[name = tensor<string, []>("op_15440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15440_cast = slice_by_index(begin = var_15440_begin_0, end = var_15440_end_0, end_mask = var_15440_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15440_cast")];
            tensor<int32, [4]> var_15441_begin_0 = const()[name = tensor<string, []>("op_15441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15441_end_0 = const()[name = tensor<string, []>("op_15441_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15441_end_mask_0 = const()[name = tensor<string, []>("op_15441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15441_cast = slice_by_index(begin = var_15441_begin_0, end = var_15441_end_0, end_mask = var_15441_end_mask_0, x = var_15411_cast)[name = tensor<string, []>("op_15441_cast")];
            tensor<int32, [4]> var_15442_begin_0 = const()[name = tensor<string, []>("op_15442_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15442_end_0 = const()[name = tensor<string, []>("op_15442_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15442_end_mask_0 = const()[name = tensor<string, []>("op_15442_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15442_cast = slice_by_index(begin = var_15442_begin_0, end = var_15442_end_0, end_mask = var_15442_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15442_cast")];
            tensor<int32, [4]> var_15443_begin_0 = const()[name = tensor<string, []>("op_15443_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15443_end_0 = const()[name = tensor<string, []>("op_15443_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15443_end_mask_0 = const()[name = tensor<string, []>("op_15443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15443_cast = slice_by_index(begin = var_15443_begin_0, end = var_15443_end_0, end_mask = var_15443_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15443_cast")];
            tensor<int32, [4]> var_15444_begin_0 = const()[name = tensor<string, []>("op_15444_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15444_end_0 = const()[name = tensor<string, []>("op_15444_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15444_end_mask_0 = const()[name = tensor<string, []>("op_15444_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15444_cast = slice_by_index(begin = var_15444_begin_0, end = var_15444_end_0, end_mask = var_15444_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15444_cast")];
            tensor<int32, [4]> var_15445_begin_0 = const()[name = tensor<string, []>("op_15445_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15445_end_0 = const()[name = tensor<string, []>("op_15445_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15445_end_mask_0 = const()[name = tensor<string, []>("op_15445_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15445_cast = slice_by_index(begin = var_15445_begin_0, end = var_15445_end_0, end_mask = var_15445_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15445_cast")];
            tensor<int32, [4]> var_15446_begin_0 = const()[name = tensor<string, []>("op_15446_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15446_end_0 = const()[name = tensor<string, []>("op_15446_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15446_end_mask_0 = const()[name = tensor<string, []>("op_15446_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15446_cast = slice_by_index(begin = var_15446_begin_0, end = var_15446_end_0, end_mask = var_15446_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15446_cast")];
            tensor<int32, [4]> var_15447_begin_0 = const()[name = tensor<string, []>("op_15447_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15447_end_0 = const()[name = tensor<string, []>("op_15447_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15447_end_mask_0 = const()[name = tensor<string, []>("op_15447_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15447_cast = slice_by_index(begin = var_15447_begin_0, end = var_15447_end_0, end_mask = var_15447_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15447_cast")];
            tensor<int32, [4]> var_15448_begin_0 = const()[name = tensor<string, []>("op_15448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15448_end_0 = const()[name = tensor<string, []>("op_15448_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15448_end_mask_0 = const()[name = tensor<string, []>("op_15448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15448_cast = slice_by_index(begin = var_15448_begin_0, end = var_15448_end_0, end_mask = var_15448_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15448_cast")];
            tensor<int32, [4]> var_15449_begin_0 = const()[name = tensor<string, []>("op_15449_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15449_end_0 = const()[name = tensor<string, []>("op_15449_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15449_end_mask_0 = const()[name = tensor<string, []>("op_15449_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15449_cast = slice_by_index(begin = var_15449_begin_0, end = var_15449_end_0, end_mask = var_15449_end_mask_0, x = var_15415_cast)[name = tensor<string, []>("op_15449_cast")];
            tensor<int32, [4]> var_15450_begin_0 = const()[name = tensor<string, []>("op_15450_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15450_end_0 = const()[name = tensor<string, []>("op_15450_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15450_end_mask_0 = const()[name = tensor<string, []>("op_15450_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15450_cast = slice_by_index(begin = var_15450_begin_0, end = var_15450_end_0, end_mask = var_15450_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15450_cast")];
            tensor<int32, [4]> var_15451_begin_0 = const()[name = tensor<string, []>("op_15451_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15451_end_0 = const()[name = tensor<string, []>("op_15451_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15451_end_mask_0 = const()[name = tensor<string, []>("op_15451_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15451_cast = slice_by_index(begin = var_15451_begin_0, end = var_15451_end_0, end_mask = var_15451_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15451_cast")];
            tensor<int32, [4]> var_15452_begin_0 = const()[name = tensor<string, []>("op_15452_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15452_end_0 = const()[name = tensor<string, []>("op_15452_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15452_end_mask_0 = const()[name = tensor<string, []>("op_15452_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15452_cast = slice_by_index(begin = var_15452_begin_0, end = var_15452_end_0, end_mask = var_15452_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15452_cast")];
            tensor<int32, [4]> var_15453_begin_0 = const()[name = tensor<string, []>("op_15453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15453_end_0 = const()[name = tensor<string, []>("op_15453_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15453_end_mask_0 = const()[name = tensor<string, []>("op_15453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15453_cast = slice_by_index(begin = var_15453_begin_0, end = var_15453_end_0, end_mask = var_15453_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15453_cast")];
            tensor<int32, [4]> var_15454_begin_0 = const()[name = tensor<string, []>("op_15454_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15454_end_0 = const()[name = tensor<string, []>("op_15454_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15454_end_mask_0 = const()[name = tensor<string, []>("op_15454_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15454_cast = slice_by_index(begin = var_15454_begin_0, end = var_15454_end_0, end_mask = var_15454_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15454_cast")];
            tensor<int32, [4]> var_15455_begin_0 = const()[name = tensor<string, []>("op_15455_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15455_end_0 = const()[name = tensor<string, []>("op_15455_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15455_end_mask_0 = const()[name = tensor<string, []>("op_15455_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15455_cast = slice_by_index(begin = var_15455_begin_0, end = var_15455_end_0, end_mask = var_15455_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15455_cast")];
            tensor<int32, [4]> var_15456_begin_0 = const()[name = tensor<string, []>("op_15456_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15456_end_0 = const()[name = tensor<string, []>("op_15456_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15456_end_mask_0 = const()[name = tensor<string, []>("op_15456_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15456_cast = slice_by_index(begin = var_15456_begin_0, end = var_15456_end_0, end_mask = var_15456_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15456_cast")];
            tensor<int32, [4]> var_15457_begin_0 = const()[name = tensor<string, []>("op_15457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15457_end_0 = const()[name = tensor<string, []>("op_15457_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15457_end_mask_0 = const()[name = tensor<string, []>("op_15457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15457_cast = slice_by_index(begin = var_15457_begin_0, end = var_15457_end_0, end_mask = var_15457_end_mask_0, x = var_15419_cast)[name = tensor<string, []>("op_15457_cast")];
            tensor<int32, [4]> var_15458_begin_0 = const()[name = tensor<string, []>("op_15458_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15458_end_0 = const()[name = tensor<string, []>("op_15458_end_0"), val = tensor<int32, [4]>([2, 64, 1, 512])];
            tensor<bool, [4]> var_15458_end_mask_0 = const()[name = tensor<string, []>("op_15458_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15458_cast = slice_by_index(begin = var_15458_begin_0, end = var_15458_end_0, end_mask = var_15458_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15458_cast")];
            tensor<int32, [4]> var_15459_begin_0 = const()[name = tensor<string, []>("op_15459_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 512])];
            tensor<int32, [4]> var_15459_end_0 = const()[name = tensor<string, []>("op_15459_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1024])];
            tensor<bool, [4]> var_15459_end_mask_0 = const()[name = tensor<string, []>("op_15459_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15459_cast = slice_by_index(begin = var_15459_begin_0, end = var_15459_end_0, end_mask = var_15459_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15459_cast")];
            tensor<int32, [4]> var_15460_begin_0 = const()[name = tensor<string, []>("op_15460_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1024])];
            tensor<int32, [4]> var_15460_end_0 = const()[name = tensor<string, []>("op_15460_end_0"), val = tensor<int32, [4]>([2, 64, 1, 1536])];
            tensor<bool, [4]> var_15460_end_mask_0 = const()[name = tensor<string, []>("op_15460_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15460_cast = slice_by_index(begin = var_15460_begin_0, end = var_15460_end_0, end_mask = var_15460_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15460_cast")];
            tensor<int32, [4]> var_15461_begin_0 = const()[name = tensor<string, []>("op_15461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1536])];
            tensor<int32, [4]> var_15461_end_0 = const()[name = tensor<string, []>("op_15461_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2048])];
            tensor<bool, [4]> var_15461_end_mask_0 = const()[name = tensor<string, []>("op_15461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15461_cast = slice_by_index(begin = var_15461_begin_0, end = var_15461_end_0, end_mask = var_15461_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15461_cast")];
            tensor<int32, [4]> var_15462_begin_0 = const()[name = tensor<string, []>("op_15462_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2048])];
            tensor<int32, [4]> var_15462_end_0 = const()[name = tensor<string, []>("op_15462_end_0"), val = tensor<int32, [4]>([2, 64, 1, 2560])];
            tensor<bool, [4]> var_15462_end_mask_0 = const()[name = tensor<string, []>("op_15462_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15462_cast = slice_by_index(begin = var_15462_begin_0, end = var_15462_end_0, end_mask = var_15462_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15462_cast")];
            tensor<int32, [4]> var_15463_begin_0 = const()[name = tensor<string, []>("op_15463_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 2560])];
            tensor<int32, [4]> var_15463_end_0 = const()[name = tensor<string, []>("op_15463_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3072])];
            tensor<bool, [4]> var_15463_end_mask_0 = const()[name = tensor<string, []>("op_15463_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15463_cast = slice_by_index(begin = var_15463_begin_0, end = var_15463_end_0, end_mask = var_15463_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15463_cast")];
            tensor<int32, [4]> var_15464_begin_0 = const()[name = tensor<string, []>("op_15464_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3072])];
            tensor<int32, [4]> var_15464_end_0 = const()[name = tensor<string, []>("op_15464_end_0"), val = tensor<int32, [4]>([2, 64, 1, 3584])];
            tensor<bool, [4]> var_15464_end_mask_0 = const()[name = tensor<string, []>("op_15464_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15464_cast = slice_by_index(begin = var_15464_begin_0, end = var_15464_end_0, end_mask = var_15464_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15464_cast")];
            tensor<int32, [4]> var_15465_begin_0 = const()[name = tensor<string, []>("op_15465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 3584])];
            tensor<int32, [4]> var_15465_end_0 = const()[name = tensor<string, []>("op_15465_end_0"), val = tensor<int32, [4]>([2, 64, 1, 4096])];
            tensor<bool, [4]> var_15465_end_mask_0 = const()[name = tensor<string, []>("op_15465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 64, 1, 512]> var_15465_cast = slice_by_index(begin = var_15465_begin_0, end = var_15465_end_0, end_mask = var_15465_end_mask_0, x = var_15423_cast)[name = tensor<string, []>("op_15465_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_15470_begin_0 = const()[name = tensor<string, []>("op_15470_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15470_end_0 = const()[name = tensor<string, []>("op_15470_end_0"), val = tensor<int32, [4]>([2, 77, 1, 64])];
            tensor<bool, [4]> var_15470_end_mask_0 = const()[name = tensor<string, []>("op_15470_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, 64]> var_15470_cast = slice_by_index(begin = var_15470_begin_0, end = var_15470_end_0, end_mask = var_15470_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_15470_cast")];
            tensor<int32, [4]> var_15474_begin_0 = const()[name = tensor<string, []>("op_15474_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_15474_end_0 = const()[name = tensor<string, []>("op_15474_end_0"), val = tensor<int32, [4]>([2, 77, 1, 128])];
            tensor<bool, [4]> var_15474_end_mask_0 = const()[name = tensor<string, []>("op_15474_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15474_cast = slice_by_index(begin = var_15474_begin_0, end = var_15474_end_0, end_mask = var_15474_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_15474_cast")];
            tensor<int32, [4]> var_15478_begin_0 = const()[name = tensor<string, []>("op_15478_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_15478_end_0 = const()[name = tensor<string, []>("op_15478_end_0"), val = tensor<int32, [4]>([2, 77, 1, 192])];
            tensor<bool, [4]> var_15478_end_mask_0 = const()[name = tensor<string, []>("op_15478_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15478_cast = slice_by_index(begin = var_15478_begin_0, end = var_15478_end_0, end_mask = var_15478_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_15478_cast")];
            tensor<int32, [4]> var_15482_begin_0 = const()[name = tensor<string, []>("op_15482_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_15482_end_0 = const()[name = tensor<string, []>("op_15482_end_0"), val = tensor<int32, [4]>([2, 77, 1, 256])];
            tensor<bool, [4]> var_15482_end_mask_0 = const()[name = tensor<string, []>("op_15482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15482_cast = slice_by_index(begin = var_15482_begin_0, end = var_15482_end_0, end_mask = var_15482_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_15482_cast")];
            tensor<int32, [4]> var_15486_begin_0 = const()[name = tensor<string, []>("op_15486_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_15486_end_0 = const()[name = tensor<string, []>("op_15486_end_0"), val = tensor<int32, [4]>([2, 77, 1, 320])];
            tensor<bool, [4]> var_15486_end_mask_0 = const()[name = tensor<string, []>("op_15486_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [2, 77, 1, 64]> var_15486_cast = slice_by_index(begin = var_15486_begin_0, end = var_15486_end_0, end_mask = var_15486_end_mask_0, x = transpose_0)[name = tensor<string, []>("op_15486_cast")];
            tensor<int32, [4]> var_15488_begin_0 = const()[name = tensor<string, []>("op_15488_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_15488_end_0 = const()[name = tensor<string, []>("op_15488_end_0"), val = tensor<int32, [4]>([2, 64, 1, 77])];
            tensor<bool, [4]> var_15488_end_mask_0 = const()[name = tensor<string, []>("op_15488_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15488_cast = slice_by_index(begin = var_15488_begin_0, end = var_15488_end_0, end_mask = var_15488_end_mask_0, x = v_cast)[name = tensor<string, []>("op_15488_cast")];
            tensor<int32, [4]> var_15492_begin_0 = const()[name = tensor<string, []>("op_15492_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_15492_end_0 = const()[name = tensor<string, []>("op_15492_end_0"), val = tensor<int32, [4]>([2, 128, 1, 77])];
            tensor<bool, [4]> var_15492_end_mask_0 = const()[name = tensor<string, []>("op_15492_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15492_cast = slice_by_index(begin = var_15492_begin_0, end = var_15492_end_0, end_mask = var_15492_end_mask_0, x = v_cast)[name = tensor<string, []>("op_15492_cast")];
            tensor<int32, [4]> var_15496_begin_0 = const()[name = tensor<string, []>("op_15496_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_15496_end_0 = const()[name = tensor<string, []>("op_15496_end_0"), val = tensor<int32, [4]>([2, 192, 1, 77])];
            tensor<bool, [4]> var_15496_end_mask_0 = const()[name = tensor<string, []>("op_15496_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15496_cast = slice_by_index(begin = var_15496_begin_0, end = var_15496_end_0, end_mask = var_15496_end_mask_0, x = v_cast)[name = tensor<string, []>("op_15496_cast")];
            tensor<int32, [4]> var_15500_begin_0 = const()[name = tensor<string, []>("op_15500_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_15500_end_0 = const()[name = tensor<string, []>("op_15500_end_0"), val = tensor<int32, [4]>([2, 256, 1, 77])];
            tensor<bool, [4]> var_15500_end_mask_0 = const()[name = tensor<string, []>("op_15500_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15500_cast = slice_by_index(begin = var_15500_begin_0, end = var_15500_end_0, end_mask = var_15500_end_mask_0, x = v_cast)[name = tensor<string, []>("op_15500_cast")];
            tensor<int32, [4]> var_15504_begin_0 = const()[name = tensor<string, []>("op_15504_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_15504_end_0 = const()[name = tensor<string, []>("op_15504_end_0"), val = tensor<int32, [4]>([2, 320, 1, 77])];
            tensor<bool, [4]> var_15504_end_mask_0 = const()[name = tensor<string, []>("op_15504_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 64, 1, 77]> var_15504_cast = slice_by_index(begin = var_15504_begin_0, end = var_15504_end_0, end_mask = var_15504_end_mask_0, x = v_cast)[name = tensor<string, []>("op_15504_cast")];
            tensor<string, []> var_15508_equation_0 = const()[name = tensor<string, []>("op_15508_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15508_cast = einsum(equation = var_15508_equation_0, values = (var_15470_cast, var_15426_cast))[name = tensor<string, []>("op_15508_cast")];
            tensor<fp16, []> var_15509_to_fp16 = const()[name = tensor<string, []>("op_15509_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1121_cast = mul(x = var_15508_cast, y = var_15509_to_fp16)[name = tensor<string, []>("aw_chunk_1121_cast")];
            tensor<string, []> var_15512_equation_0 = const()[name = tensor<string, []>("op_15512_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15512_cast = einsum(equation = var_15512_equation_0, values = (var_15470_cast, var_15427_cast))[name = tensor<string, []>("op_15512_cast")];
            tensor<fp16, []> var_15513_to_fp16 = const()[name = tensor<string, []>("op_15513_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1123_cast = mul(x = var_15512_cast, y = var_15513_to_fp16)[name = tensor<string, []>("aw_chunk_1123_cast")];
            tensor<string, []> var_15516_equation_0 = const()[name = tensor<string, []>("op_15516_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15516_cast = einsum(equation = var_15516_equation_0, values = (var_15470_cast, var_15428_cast))[name = tensor<string, []>("op_15516_cast")];
            tensor<fp16, []> var_15517_to_fp16 = const()[name = tensor<string, []>("op_15517_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1125_cast = mul(x = var_15516_cast, y = var_15517_to_fp16)[name = tensor<string, []>("aw_chunk_1125_cast")];
            tensor<string, []> var_15520_equation_0 = const()[name = tensor<string, []>("op_15520_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15520_cast = einsum(equation = var_15520_equation_0, values = (var_15470_cast, var_15429_cast))[name = tensor<string, []>("op_15520_cast")];
            tensor<fp16, []> var_15521_to_fp16 = const()[name = tensor<string, []>("op_15521_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1127_cast = mul(x = var_15520_cast, y = var_15521_to_fp16)[name = tensor<string, []>("aw_chunk_1127_cast")];
            tensor<string, []> var_15524_equation_0 = const()[name = tensor<string, []>("op_15524_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15524_cast = einsum(equation = var_15524_equation_0, values = (var_15470_cast, var_15430_cast))[name = tensor<string, []>("op_15524_cast")];
            tensor<fp16, []> var_15525_to_fp16 = const()[name = tensor<string, []>("op_15525_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1129_cast = mul(x = var_15524_cast, y = var_15525_to_fp16)[name = tensor<string, []>("aw_chunk_1129_cast")];
            tensor<string, []> var_15528_equation_0 = const()[name = tensor<string, []>("op_15528_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15528_cast = einsum(equation = var_15528_equation_0, values = (var_15470_cast, var_15431_cast))[name = tensor<string, []>("op_15528_cast")];
            tensor<fp16, []> var_15529_to_fp16 = const()[name = tensor<string, []>("op_15529_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1131_cast = mul(x = var_15528_cast, y = var_15529_to_fp16)[name = tensor<string, []>("aw_chunk_1131_cast")];
            tensor<string, []> var_15532_equation_0 = const()[name = tensor<string, []>("op_15532_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15532_cast = einsum(equation = var_15532_equation_0, values = (var_15470_cast, var_15432_cast))[name = tensor<string, []>("op_15532_cast")];
            tensor<fp16, []> var_15533_to_fp16 = const()[name = tensor<string, []>("op_15533_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1133_cast = mul(x = var_15532_cast, y = var_15533_to_fp16)[name = tensor<string, []>("aw_chunk_1133_cast")];
            tensor<string, []> var_15536_equation_0 = const()[name = tensor<string, []>("op_15536_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15536_cast = einsum(equation = var_15536_equation_0, values = (var_15470_cast, var_15433_cast))[name = tensor<string, []>("op_15536_cast")];
            tensor<fp16, []> var_15537_to_fp16 = const()[name = tensor<string, []>("op_15537_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1135_cast = mul(x = var_15536_cast, y = var_15537_to_fp16)[name = tensor<string, []>("aw_chunk_1135_cast")];
            tensor<string, []> var_15540_equation_0 = const()[name = tensor<string, []>("op_15540_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15540_cast = einsum(equation = var_15540_equation_0, values = (var_15474_cast, var_15434_cast))[name = tensor<string, []>("op_15540_cast")];
            tensor<fp16, []> var_15541_to_fp16 = const()[name = tensor<string, []>("op_15541_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1137_cast = mul(x = var_15540_cast, y = var_15541_to_fp16)[name = tensor<string, []>("aw_chunk_1137_cast")];
            tensor<string, []> var_15544_equation_0 = const()[name = tensor<string, []>("op_15544_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15544_cast = einsum(equation = var_15544_equation_0, values = (var_15474_cast, var_15435_cast))[name = tensor<string, []>("op_15544_cast")];
            tensor<fp16, []> var_15545_to_fp16 = const()[name = tensor<string, []>("op_15545_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1139_cast = mul(x = var_15544_cast, y = var_15545_to_fp16)[name = tensor<string, []>("aw_chunk_1139_cast")];
            tensor<string, []> var_15548_equation_0 = const()[name = tensor<string, []>("op_15548_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15548_cast = einsum(equation = var_15548_equation_0, values = (var_15474_cast, var_15436_cast))[name = tensor<string, []>("op_15548_cast")];
            tensor<fp16, []> var_15549_to_fp16 = const()[name = tensor<string, []>("op_15549_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1141_cast = mul(x = var_15548_cast, y = var_15549_to_fp16)[name = tensor<string, []>("aw_chunk_1141_cast")];
            tensor<string, []> var_15552_equation_0 = const()[name = tensor<string, []>("op_15552_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15552_cast = einsum(equation = var_15552_equation_0, values = (var_15474_cast, var_15437_cast))[name = tensor<string, []>("op_15552_cast")];
            tensor<fp16, []> var_15553_to_fp16 = const()[name = tensor<string, []>("op_15553_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1143_cast = mul(x = var_15552_cast, y = var_15553_to_fp16)[name = tensor<string, []>("aw_chunk_1143_cast")];
            tensor<string, []> var_15556_equation_0 = const()[name = tensor<string, []>("op_15556_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15556_cast = einsum(equation = var_15556_equation_0, values = (var_15474_cast, var_15438_cast))[name = tensor<string, []>("op_15556_cast")];
            tensor<fp16, []> var_15557_to_fp16 = const()[name = tensor<string, []>("op_15557_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1145_cast = mul(x = var_15556_cast, y = var_15557_to_fp16)[name = tensor<string, []>("aw_chunk_1145_cast")];
            tensor<string, []> var_15560_equation_0 = const()[name = tensor<string, []>("op_15560_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15560_cast = einsum(equation = var_15560_equation_0, values = (var_15474_cast, var_15439_cast))[name = tensor<string, []>("op_15560_cast")];
            tensor<fp16, []> var_15561_to_fp16 = const()[name = tensor<string, []>("op_15561_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1147_cast = mul(x = var_15560_cast, y = var_15561_to_fp16)[name = tensor<string, []>("aw_chunk_1147_cast")];
            tensor<string, []> var_15564_equation_0 = const()[name = tensor<string, []>("op_15564_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15564_cast = einsum(equation = var_15564_equation_0, values = (var_15474_cast, var_15440_cast))[name = tensor<string, []>("op_15564_cast")];
            tensor<fp16, []> var_15565_to_fp16 = const()[name = tensor<string, []>("op_15565_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1149_cast = mul(x = var_15564_cast, y = var_15565_to_fp16)[name = tensor<string, []>("aw_chunk_1149_cast")];
            tensor<string, []> var_15568_equation_0 = const()[name = tensor<string, []>("op_15568_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15568_cast = einsum(equation = var_15568_equation_0, values = (var_15474_cast, var_15441_cast))[name = tensor<string, []>("op_15568_cast")];
            tensor<fp16, []> var_15569_to_fp16 = const()[name = tensor<string, []>("op_15569_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1151_cast = mul(x = var_15568_cast, y = var_15569_to_fp16)[name = tensor<string, []>("aw_chunk_1151_cast")];
            tensor<string, []> var_15572_equation_0 = const()[name = tensor<string, []>("op_15572_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15572_cast = einsum(equation = var_15572_equation_0, values = (var_15478_cast, var_15442_cast))[name = tensor<string, []>("op_15572_cast")];
            tensor<fp16, []> var_15573_to_fp16 = const()[name = tensor<string, []>("op_15573_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1153_cast = mul(x = var_15572_cast, y = var_15573_to_fp16)[name = tensor<string, []>("aw_chunk_1153_cast")];
            tensor<string, []> var_15576_equation_0 = const()[name = tensor<string, []>("op_15576_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15576_cast = einsum(equation = var_15576_equation_0, values = (var_15478_cast, var_15443_cast))[name = tensor<string, []>("op_15576_cast")];
            tensor<fp16, []> var_15577_to_fp16 = const()[name = tensor<string, []>("op_15577_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1155_cast = mul(x = var_15576_cast, y = var_15577_to_fp16)[name = tensor<string, []>("aw_chunk_1155_cast")];
            tensor<string, []> var_15580_equation_0 = const()[name = tensor<string, []>("op_15580_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15580_cast = einsum(equation = var_15580_equation_0, values = (var_15478_cast, var_15444_cast))[name = tensor<string, []>("op_15580_cast")];
            tensor<fp16, []> var_15581_to_fp16 = const()[name = tensor<string, []>("op_15581_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1157_cast = mul(x = var_15580_cast, y = var_15581_to_fp16)[name = tensor<string, []>("aw_chunk_1157_cast")];
            tensor<string, []> var_15584_equation_0 = const()[name = tensor<string, []>("op_15584_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15584_cast = einsum(equation = var_15584_equation_0, values = (var_15478_cast, var_15445_cast))[name = tensor<string, []>("op_15584_cast")];
            tensor<fp16, []> var_15585_to_fp16 = const()[name = tensor<string, []>("op_15585_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1159_cast = mul(x = var_15584_cast, y = var_15585_to_fp16)[name = tensor<string, []>("aw_chunk_1159_cast")];
            tensor<string, []> var_15588_equation_0 = const()[name = tensor<string, []>("op_15588_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15588_cast = einsum(equation = var_15588_equation_0, values = (var_15478_cast, var_15446_cast))[name = tensor<string, []>("op_15588_cast")];
            tensor<fp16, []> var_15589_to_fp16 = const()[name = tensor<string, []>("op_15589_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1161_cast = mul(x = var_15588_cast, y = var_15589_to_fp16)[name = tensor<string, []>("aw_chunk_1161_cast")];
            tensor<string, []> var_15592_equation_0 = const()[name = tensor<string, []>("op_15592_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15592_cast = einsum(equation = var_15592_equation_0, values = (var_15478_cast, var_15447_cast))[name = tensor<string, []>("op_15592_cast")];
            tensor<fp16, []> var_15593_to_fp16 = const()[name = tensor<string, []>("op_15593_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1163_cast = mul(x = var_15592_cast, y = var_15593_to_fp16)[name = tensor<string, []>("aw_chunk_1163_cast")];
            tensor<string, []> var_15596_equation_0 = const()[name = tensor<string, []>("op_15596_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15596_cast = einsum(equation = var_15596_equation_0, values = (var_15478_cast, var_15448_cast))[name = tensor<string, []>("op_15596_cast")];
            tensor<fp16, []> var_15597_to_fp16 = const()[name = tensor<string, []>("op_15597_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1165_cast = mul(x = var_15596_cast, y = var_15597_to_fp16)[name = tensor<string, []>("aw_chunk_1165_cast")];
            tensor<string, []> var_15600_equation_0 = const()[name = tensor<string, []>("op_15600_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15600_cast = einsum(equation = var_15600_equation_0, values = (var_15478_cast, var_15449_cast))[name = tensor<string, []>("op_15600_cast")];
            tensor<fp16, []> var_15601_to_fp16 = const()[name = tensor<string, []>("op_15601_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1167_cast = mul(x = var_15600_cast, y = var_15601_to_fp16)[name = tensor<string, []>("aw_chunk_1167_cast")];
            tensor<string, []> var_15604_equation_0 = const()[name = tensor<string, []>("op_15604_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15604_cast = einsum(equation = var_15604_equation_0, values = (var_15482_cast, var_15450_cast))[name = tensor<string, []>("op_15604_cast")];
            tensor<fp16, []> var_15605_to_fp16 = const()[name = tensor<string, []>("op_15605_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1169_cast = mul(x = var_15604_cast, y = var_15605_to_fp16)[name = tensor<string, []>("aw_chunk_1169_cast")];
            tensor<string, []> var_15608_equation_0 = const()[name = tensor<string, []>("op_15608_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15608_cast = einsum(equation = var_15608_equation_0, values = (var_15482_cast, var_15451_cast))[name = tensor<string, []>("op_15608_cast")];
            tensor<fp16, []> var_15609_to_fp16 = const()[name = tensor<string, []>("op_15609_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1171_cast = mul(x = var_15608_cast, y = var_15609_to_fp16)[name = tensor<string, []>("aw_chunk_1171_cast")];
            tensor<string, []> var_15612_equation_0 = const()[name = tensor<string, []>("op_15612_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15612_cast = einsum(equation = var_15612_equation_0, values = (var_15482_cast, var_15452_cast))[name = tensor<string, []>("op_15612_cast")];
            tensor<fp16, []> var_15613_to_fp16 = const()[name = tensor<string, []>("op_15613_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1173_cast = mul(x = var_15612_cast, y = var_15613_to_fp16)[name = tensor<string, []>("aw_chunk_1173_cast")];
            tensor<string, []> var_15616_equation_0 = const()[name = tensor<string, []>("op_15616_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15616_cast = einsum(equation = var_15616_equation_0, values = (var_15482_cast, var_15453_cast))[name = tensor<string, []>("op_15616_cast")];
            tensor<fp16, []> var_15617_to_fp16 = const()[name = tensor<string, []>("op_15617_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1175_cast = mul(x = var_15616_cast, y = var_15617_to_fp16)[name = tensor<string, []>("aw_chunk_1175_cast")];
            tensor<string, []> var_15620_equation_0 = const()[name = tensor<string, []>("op_15620_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15620_cast = einsum(equation = var_15620_equation_0, values = (var_15482_cast, var_15454_cast))[name = tensor<string, []>("op_15620_cast")];
            tensor<fp16, []> var_15621_to_fp16 = const()[name = tensor<string, []>("op_15621_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1177_cast = mul(x = var_15620_cast, y = var_15621_to_fp16)[name = tensor<string, []>("aw_chunk_1177_cast")];
            tensor<string, []> var_15624_equation_0 = const()[name = tensor<string, []>("op_15624_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15624_cast = einsum(equation = var_15624_equation_0, values = (var_15482_cast, var_15455_cast))[name = tensor<string, []>("op_15624_cast")];
            tensor<fp16, []> var_15625_to_fp16 = const()[name = tensor<string, []>("op_15625_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1179_cast = mul(x = var_15624_cast, y = var_15625_to_fp16)[name = tensor<string, []>("aw_chunk_1179_cast")];
            tensor<string, []> var_15628_equation_0 = const()[name = tensor<string, []>("op_15628_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15628_cast = einsum(equation = var_15628_equation_0, values = (var_15482_cast, var_15456_cast))[name = tensor<string, []>("op_15628_cast")];
            tensor<fp16, []> var_15629_to_fp16 = const()[name = tensor<string, []>("op_15629_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1181_cast = mul(x = var_15628_cast, y = var_15629_to_fp16)[name = tensor<string, []>("aw_chunk_1181_cast")];
            tensor<string, []> var_15632_equation_0 = const()[name = tensor<string, []>("op_15632_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15632_cast = einsum(equation = var_15632_equation_0, values = (var_15482_cast, var_15457_cast))[name = tensor<string, []>("op_15632_cast")];
            tensor<fp16, []> var_15633_to_fp16 = const()[name = tensor<string, []>("op_15633_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1183_cast = mul(x = var_15632_cast, y = var_15633_to_fp16)[name = tensor<string, []>("aw_chunk_1183_cast")];
            tensor<string, []> var_15636_equation_0 = const()[name = tensor<string, []>("op_15636_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15636_cast = einsum(equation = var_15636_equation_0, values = (var_15486_cast, var_15458_cast))[name = tensor<string, []>("op_15636_cast")];
            tensor<fp16, []> var_15637_to_fp16 = const()[name = tensor<string, []>("op_15637_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1185_cast = mul(x = var_15636_cast, y = var_15637_to_fp16)[name = tensor<string, []>("aw_chunk_1185_cast")];
            tensor<string, []> var_15640_equation_0 = const()[name = tensor<string, []>("op_15640_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15640_cast = einsum(equation = var_15640_equation_0, values = (var_15486_cast, var_15459_cast))[name = tensor<string, []>("op_15640_cast")];
            tensor<fp16, []> var_15641_to_fp16 = const()[name = tensor<string, []>("op_15641_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1187_cast = mul(x = var_15640_cast, y = var_15641_to_fp16)[name = tensor<string, []>("aw_chunk_1187_cast")];
            tensor<string, []> var_15644_equation_0 = const()[name = tensor<string, []>("op_15644_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15644_cast = einsum(equation = var_15644_equation_0, values = (var_15486_cast, var_15460_cast))[name = tensor<string, []>("op_15644_cast")];
            tensor<fp16, []> var_15645_to_fp16 = const()[name = tensor<string, []>("op_15645_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1189_cast = mul(x = var_15644_cast, y = var_15645_to_fp16)[name = tensor<string, []>("aw_chunk_1189_cast")];
            tensor<string, []> var_15648_equation_0 = const()[name = tensor<string, []>("op_15648_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15648_cast = einsum(equation = var_15648_equation_0, values = (var_15486_cast, var_15461_cast))[name = tensor<string, []>("op_15648_cast")];
            tensor<fp16, []> var_15649_to_fp16 = const()[name = tensor<string, []>("op_15649_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1191_cast = mul(x = var_15648_cast, y = var_15649_to_fp16)[name = tensor<string, []>("aw_chunk_1191_cast")];
            tensor<string, []> var_15652_equation_0 = const()[name = tensor<string, []>("op_15652_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15652_cast = einsum(equation = var_15652_equation_0, values = (var_15486_cast, var_15462_cast))[name = tensor<string, []>("op_15652_cast")];
            tensor<fp16, []> var_15653_to_fp16 = const()[name = tensor<string, []>("op_15653_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1193_cast = mul(x = var_15652_cast, y = var_15653_to_fp16)[name = tensor<string, []>("aw_chunk_1193_cast")];
            tensor<string, []> var_15656_equation_0 = const()[name = tensor<string, []>("op_15656_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15656_cast = einsum(equation = var_15656_equation_0, values = (var_15486_cast, var_15463_cast))[name = tensor<string, []>("op_15656_cast")];
            tensor<fp16, []> var_15657_to_fp16 = const()[name = tensor<string, []>("op_15657_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1195_cast = mul(x = var_15656_cast, y = var_15657_to_fp16)[name = tensor<string, []>("aw_chunk_1195_cast")];
            tensor<string, []> var_15660_equation_0 = const()[name = tensor<string, []>("op_15660_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15660_cast = einsum(equation = var_15660_equation_0, values = (var_15486_cast, var_15464_cast))[name = tensor<string, []>("op_15660_cast")];
            tensor<fp16, []> var_15661_to_fp16 = const()[name = tensor<string, []>("op_15661_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_1197_cast = mul(x = var_15660_cast, y = var_15661_to_fp16)[name = tensor<string, []>("aw_chunk_1197_cast")];
            tensor<string, []> var_15664_equation_0 = const()[name = tensor<string, []>("op_15664_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")];
            tensor<fp16, [2, 77, 1, 512]> var_15664_cast = einsum(equation = var_15664_equation_0, values = (var_15486_cast, var_15465_cast))[name = tensor<string, []>("op_15664_cast")];
            tensor<fp16, []> var_15665_to_fp16 = const()[name = tensor<string, []>("op_15665_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [2, 77, 1, 512]> aw_chunk_cast = mul(x = var_15664_cast, y = var_15665_to_fp16)[name = tensor<string, []>("aw_chunk_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15667_cast = softmax(axis = var_12807, x = aw_chunk_1121_cast)[name = tensor<string, []>("op_15667_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15668_cast = softmax(axis = var_12807, x = aw_chunk_1123_cast)[name = tensor<string, []>("op_15668_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15669_cast = softmax(axis = var_12807, x = aw_chunk_1125_cast)[name = tensor<string, []>("op_15669_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15670_cast = softmax(axis = var_12807, x = aw_chunk_1127_cast)[name = tensor<string, []>("op_15670_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15671_cast = softmax(axis = var_12807, x = aw_chunk_1129_cast)[name = tensor<string, []>("op_15671_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15672_cast = softmax(axis = var_12807, x = aw_chunk_1131_cast)[name = tensor<string, []>("op_15672_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15673_cast = softmax(axis = var_12807, x = aw_chunk_1133_cast)[name = tensor<string, []>("op_15673_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15674_cast = softmax(axis = var_12807, x = aw_chunk_1135_cast)[name = tensor<string, []>("op_15674_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15675_cast = softmax(axis = var_12807, x = aw_chunk_1137_cast)[name = tensor<string, []>("op_15675_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15676_cast = softmax(axis = var_12807, x = aw_chunk_1139_cast)[name = tensor<string, []>("op_15676_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15677_cast = softmax(axis = var_12807, x = aw_chunk_1141_cast)[name = tensor<string, []>("op_15677_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15678_cast = softmax(axis = var_12807, x = aw_chunk_1143_cast)[name = tensor<string, []>("op_15678_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15679_cast = softmax(axis = var_12807, x = aw_chunk_1145_cast)[name = tensor<string, []>("op_15679_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15680_cast = softmax(axis = var_12807, x = aw_chunk_1147_cast)[name = tensor<string, []>("op_15680_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15681_cast = softmax(axis = var_12807, x = aw_chunk_1149_cast)[name = tensor<string, []>("op_15681_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15682_cast = softmax(axis = var_12807, x = aw_chunk_1151_cast)[name = tensor<string, []>("op_15682_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15683_cast = softmax(axis = var_12807, x = aw_chunk_1153_cast)[name = tensor<string, []>("op_15683_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15684_cast = softmax(axis = var_12807, x = aw_chunk_1155_cast)[name = tensor<string, []>("op_15684_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15685_cast = softmax(axis = var_12807, x = aw_chunk_1157_cast)[name = tensor<string, []>("op_15685_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15686_cast = softmax(axis = var_12807, x = aw_chunk_1159_cast)[name = tensor<string, []>("op_15686_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15687_cast = softmax(axis = var_12807, x = aw_chunk_1161_cast)[name = tensor<string, []>("op_15687_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15688_cast = softmax(axis = var_12807, x = aw_chunk_1163_cast)[name = tensor<string, []>("op_15688_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15689_cast = softmax(axis = var_12807, x = aw_chunk_1165_cast)[name = tensor<string, []>("op_15689_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15690_cast = softmax(axis = var_12807, x = aw_chunk_1167_cast)[name = tensor<string, []>("op_15690_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15691_cast = softmax(axis = var_12807, x = aw_chunk_1169_cast)[name = tensor<string, []>("op_15691_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15692_cast = softmax(axis = var_12807, x = aw_chunk_1171_cast)[name = tensor<string, []>("op_15692_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15693_cast = softmax(axis = var_12807, x = aw_chunk_1173_cast)[name = tensor<string, []>("op_15693_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15694_cast = softmax(axis = var_12807, x = aw_chunk_1175_cast)[name = tensor<string, []>("op_15694_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15695_cast = softmax(axis = var_12807, x = aw_chunk_1177_cast)[name = tensor<string, []>("op_15695_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15696_cast = softmax(axis = var_12807, x = aw_chunk_1179_cast)[name = tensor<string, []>("op_15696_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15697_cast = softmax(axis = var_12807, x = aw_chunk_1181_cast)[name = tensor<string, []>("op_15697_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15698_cast = softmax(axis = var_12807, x = aw_chunk_1183_cast)[name = tensor<string, []>("op_15698_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15699_cast = softmax(axis = var_12807, x = aw_chunk_1185_cast)[name = tensor<string, []>("op_15699_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15700_cast = softmax(axis = var_12807, x = aw_chunk_1187_cast)[name = tensor<string, []>("op_15700_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15701_cast = softmax(axis = var_12807, x = aw_chunk_1189_cast)[name = tensor<string, []>("op_15701_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15702_cast = softmax(axis = var_12807, x = aw_chunk_1191_cast)[name = tensor<string, []>("op_15702_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15703_cast = softmax(axis = var_12807, x = aw_chunk_1193_cast)[name = tensor<string, []>("op_15703_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15704_cast = softmax(axis = var_12807, x = aw_chunk_1195_cast)[name = tensor<string, []>("op_15704_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15705_cast = softmax(axis = var_12807, x = aw_chunk_1197_cast)[name = tensor<string, []>("op_15705_cast")];
            tensor<fp16, [2, 77, 1, 512]> var_15706_cast = softmax(axis = var_12807, x = aw_chunk_cast)[name = tensor<string, []>("op_15706_cast")];
            tensor<string, []> var_15708_equation_0 = const()[name = tensor<string, []>("op_15708_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15708_cast = einsum(equation = var_15708_equation_0, values = (var_15488_cast, var_15667_cast))[name = tensor<string, []>("op_15708_cast")];
            tensor<string, []> var_15710_equation_0 = const()[name = tensor<string, []>("op_15710_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15710_cast = einsum(equation = var_15710_equation_0, values = (var_15488_cast, var_15668_cast))[name = tensor<string, []>("op_15710_cast")];
            tensor<string, []> var_15712_equation_0 = const()[name = tensor<string, []>("op_15712_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15712_cast = einsum(equation = var_15712_equation_0, values = (var_15488_cast, var_15669_cast))[name = tensor<string, []>("op_15712_cast")];
            tensor<string, []> var_15714_equation_0 = const()[name = tensor<string, []>("op_15714_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15714_cast = einsum(equation = var_15714_equation_0, values = (var_15488_cast, var_15670_cast))[name = tensor<string, []>("op_15714_cast")];
            tensor<string, []> var_15716_equation_0 = const()[name = tensor<string, []>("op_15716_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15716_cast = einsum(equation = var_15716_equation_0, values = (var_15488_cast, var_15671_cast))[name = tensor<string, []>("op_15716_cast")];
            tensor<string, []> var_15718_equation_0 = const()[name = tensor<string, []>("op_15718_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15718_cast = einsum(equation = var_15718_equation_0, values = (var_15488_cast, var_15672_cast))[name = tensor<string, []>("op_15718_cast")];
            tensor<string, []> var_15720_equation_0 = const()[name = tensor<string, []>("op_15720_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15720_cast = einsum(equation = var_15720_equation_0, values = (var_15488_cast, var_15673_cast))[name = tensor<string, []>("op_15720_cast")];
            tensor<string, []> var_15722_equation_0 = const()[name = tensor<string, []>("op_15722_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15722_cast = einsum(equation = var_15722_equation_0, values = (var_15488_cast, var_15674_cast))[name = tensor<string, []>("op_15722_cast")];
            tensor<string, []> var_15724_equation_0 = const()[name = tensor<string, []>("op_15724_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15724_cast = einsum(equation = var_15724_equation_0, values = (var_15492_cast, var_15675_cast))[name = tensor<string, []>("op_15724_cast")];
            tensor<string, []> var_15726_equation_0 = const()[name = tensor<string, []>("op_15726_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15726_cast = einsum(equation = var_15726_equation_0, values = (var_15492_cast, var_15676_cast))[name = tensor<string, []>("op_15726_cast")];
            tensor<string, []> var_15728_equation_0 = const()[name = tensor<string, []>("op_15728_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15728_cast = einsum(equation = var_15728_equation_0, values = (var_15492_cast, var_15677_cast))[name = tensor<string, []>("op_15728_cast")];
            tensor<string, []> var_15730_equation_0 = const()[name = tensor<string, []>("op_15730_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15730_cast = einsum(equation = var_15730_equation_0, values = (var_15492_cast, var_15678_cast))[name = tensor<string, []>("op_15730_cast")];
            tensor<string, []> var_15732_equation_0 = const()[name = tensor<string, []>("op_15732_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15732_cast = einsum(equation = var_15732_equation_0, values = (var_15492_cast, var_15679_cast))[name = tensor<string, []>("op_15732_cast")];
            tensor<string, []> var_15734_equation_0 = const()[name = tensor<string, []>("op_15734_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15734_cast = einsum(equation = var_15734_equation_0, values = (var_15492_cast, var_15680_cast))[name = tensor<string, []>("op_15734_cast")];
            tensor<string, []> var_15736_equation_0 = const()[name = tensor<string, []>("op_15736_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15736_cast = einsum(equation = var_15736_equation_0, values = (var_15492_cast, var_15681_cast))[name = tensor<string, []>("op_15736_cast")];
            tensor<string, []> var_15738_equation_0 = const()[name = tensor<string, []>("op_15738_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15738_cast = einsum(equation = var_15738_equation_0, values = (var_15492_cast, var_15682_cast))[name = tensor<string, []>("op_15738_cast")];
            tensor<string, []> var_15740_equation_0 = const()[name = tensor<string, []>("op_15740_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15740_cast = einsum(equation = var_15740_equation_0, values = (var_15496_cast, var_15683_cast))[name = tensor<string, []>("op_15740_cast")];
            tensor<string, []> var_15742_equation_0 = const()[name = tensor<string, []>("op_15742_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15742_cast = einsum(equation = var_15742_equation_0, values = (var_15496_cast, var_15684_cast))[name = tensor<string, []>("op_15742_cast")];
            tensor<string, []> var_15744_equation_0 = const()[name = tensor<string, []>("op_15744_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15744_cast = einsum(equation = var_15744_equation_0, values = (var_15496_cast, var_15685_cast))[name = tensor<string, []>("op_15744_cast")];
            tensor<string, []> var_15746_equation_0 = const()[name = tensor<string, []>("op_15746_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15746_cast = einsum(equation = var_15746_equation_0, values = (var_15496_cast, var_15686_cast))[name = tensor<string, []>("op_15746_cast")];
            tensor<string, []> var_15748_equation_0 = const()[name = tensor<string, []>("op_15748_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15748_cast = einsum(equation = var_15748_equation_0, values = (var_15496_cast, var_15687_cast))[name = tensor<string, []>("op_15748_cast")];
            tensor<string, []> var_15750_equation_0 = const()[name = tensor<string, []>("op_15750_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15750_cast = einsum(equation = var_15750_equation_0, values = (var_15496_cast, var_15688_cast))[name = tensor<string, []>("op_15750_cast")];
            tensor<string, []> var_15752_equation_0 = const()[name = tensor<string, []>("op_15752_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15752_cast = einsum(equation = var_15752_equation_0, values = (var_15496_cast, var_15689_cast))[name = tensor<string, []>("op_15752_cast")];
            tensor<string, []> var_15754_equation_0 = const()[name = tensor<string, []>("op_15754_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15754_cast = einsum(equation = var_15754_equation_0, values = (var_15496_cast, var_15690_cast))[name = tensor<string, []>("op_15754_cast")];
            tensor<string, []> var_15756_equation_0 = const()[name = tensor<string, []>("op_15756_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15756_cast = einsum(equation = var_15756_equation_0, values = (var_15500_cast, var_15691_cast))[name = tensor<string, []>("op_15756_cast")];
            tensor<string, []> var_15758_equation_0 = const()[name = tensor<string, []>("op_15758_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15758_cast = einsum(equation = var_15758_equation_0, values = (var_15500_cast, var_15692_cast))[name = tensor<string, []>("op_15758_cast")];
            tensor<string, []> var_15760_equation_0 = const()[name = tensor<string, []>("op_15760_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15760_cast = einsum(equation = var_15760_equation_0, values = (var_15500_cast, var_15693_cast))[name = tensor<string, []>("op_15760_cast")];
            tensor<string, []> var_15762_equation_0 = const()[name = tensor<string, []>("op_15762_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15762_cast = einsum(equation = var_15762_equation_0, values = (var_15500_cast, var_15694_cast))[name = tensor<string, []>("op_15762_cast")];
            tensor<string, []> var_15764_equation_0 = const()[name = tensor<string, []>("op_15764_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15764_cast = einsum(equation = var_15764_equation_0, values = (var_15500_cast, var_15695_cast))[name = tensor<string, []>("op_15764_cast")];
            tensor<string, []> var_15766_equation_0 = const()[name = tensor<string, []>("op_15766_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15766_cast = einsum(equation = var_15766_equation_0, values = (var_15500_cast, var_15696_cast))[name = tensor<string, []>("op_15766_cast")];
            tensor<string, []> var_15768_equation_0 = const()[name = tensor<string, []>("op_15768_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15768_cast = einsum(equation = var_15768_equation_0, values = (var_15500_cast, var_15697_cast))[name = tensor<string, []>("op_15768_cast")];
            tensor<string, []> var_15770_equation_0 = const()[name = tensor<string, []>("op_15770_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15770_cast = einsum(equation = var_15770_equation_0, values = (var_15500_cast, var_15698_cast))[name = tensor<string, []>("op_15770_cast")];
            tensor<string, []> var_15772_equation_0 = const()[name = tensor<string, []>("op_15772_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15772_cast = einsum(equation = var_15772_equation_0, values = (var_15504_cast, var_15699_cast))[name = tensor<string, []>("op_15772_cast")];
            tensor<string, []> var_15774_equation_0 = const()[name = tensor<string, []>("op_15774_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15774_cast = einsum(equation = var_15774_equation_0, values = (var_15504_cast, var_15700_cast))[name = tensor<string, []>("op_15774_cast")];
            tensor<string, []> var_15776_equation_0 = const()[name = tensor<string, []>("op_15776_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15776_cast = einsum(equation = var_15776_equation_0, values = (var_15504_cast, var_15701_cast))[name = tensor<string, []>("op_15776_cast")];
            tensor<string, []> var_15778_equation_0 = const()[name = tensor<string, []>("op_15778_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15778_cast = einsum(equation = var_15778_equation_0, values = (var_15504_cast, var_15702_cast))[name = tensor<string, []>("op_15778_cast")];
            tensor<string, []> var_15780_equation_0 = const()[name = tensor<string, []>("op_15780_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15780_cast = einsum(equation = var_15780_equation_0, values = (var_15504_cast, var_15703_cast))[name = tensor<string, []>("op_15780_cast")];
            tensor<string, []> var_15782_equation_0 = const()[name = tensor<string, []>("op_15782_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15782_cast = einsum(equation = var_15782_equation_0, values = (var_15504_cast, var_15704_cast))[name = tensor<string, []>("op_15782_cast")];
            tensor<string, []> var_15784_equation_0 = const()[name = tensor<string, []>("op_15784_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15784_cast = einsum(equation = var_15784_equation_0, values = (var_15504_cast, var_15705_cast))[name = tensor<string, []>("op_15784_cast")];
            tensor<string, []> var_15786_equation_0 = const()[name = tensor<string, []>("op_15786_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")];
            tensor<fp16, [2, 64, 1, 512]> var_15786_cast = einsum(equation = var_15786_equation_0, values = (var_15504_cast, var_15706_cast))[name = tensor<string, []>("op_15786_cast")];
            tensor<bool, []> var_15788_interleave_0 = const()[name = tensor<string, []>("op_15788_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15788_cast = concat(axis = var_12782, interleave = var_15788_interleave_0, values = (var_15708_cast, var_15710_cast, var_15712_cast, var_15714_cast, var_15716_cast, var_15718_cast, var_15720_cast, var_15722_cast))[name = tensor<string, []>("op_15788_cast")];
            tensor<bool, []> var_15790_interleave_0 = const()[name = tensor<string, []>("op_15790_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15790_cast = concat(axis = var_12782, interleave = var_15790_interleave_0, values = (var_15724_cast, var_15726_cast, var_15728_cast, var_15730_cast, var_15732_cast, var_15734_cast, var_15736_cast, var_15738_cast))[name = tensor<string, []>("op_15790_cast")];
            tensor<bool, []> var_15792_interleave_0 = const()[name = tensor<string, []>("op_15792_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15792_cast = concat(axis = var_12782, interleave = var_15792_interleave_0, values = (var_15740_cast, var_15742_cast, var_15744_cast, var_15746_cast, var_15748_cast, var_15750_cast, var_15752_cast, var_15754_cast))[name = tensor<string, []>("op_15792_cast")];
            tensor<bool, []> var_15794_interleave_0 = const()[name = tensor<string, []>("op_15794_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15794_cast = concat(axis = var_12782, interleave = var_15794_interleave_0, values = (var_15756_cast, var_15758_cast, var_15760_cast, var_15762_cast, var_15764_cast, var_15766_cast, var_15768_cast, var_15770_cast))[name = tensor<string, []>("op_15794_cast")];
            tensor<bool, []> var_15796_interleave_0 = const()[name = tensor<string, []>("op_15796_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [2, 64, 1, 4096]> var_15796_cast = concat(axis = var_12782, interleave = var_15796_interleave_0, values = (var_15772_cast, var_15774_cast, var_15776_cast, var_15778_cast, var_15780_cast, var_15782_cast, var_15784_cast, var_15786_cast))[name = tensor<string, []>("op_15796_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_12807, interleave = input_519_interleave_0, values = (var_15788_cast, var_15790_cast, var_15792_cast, var_15794_cast, var_15796_cast))[name = tensor<string, []>("input_519_cast")];
            tensor<int32, [2]> var_15802 = const()[name = tensor<string, []>("op_15802"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15804 = const()[name = tensor<string, []>("op_15804"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15806_pad_type_0 = const()[name = tensor<string, []>("op_15806_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15806_pad_0 = const()[name = tensor<string, []>("op_15806_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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216734464))), lut = tensor<fp16, [4]>([-0x1.7b8p-6, -0x1.21p-9, 0x1.1c4p-9, 0x1.6e8p-6]), 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, []>(216760128)))];
            tensor<fp16, [2, 320, 1, 4096]> var_15806_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_attn2_to_out_0_bias_to_fp16, dilations = var_15804, groups = var_12807, pad = var_15806_pad_0, pad_type = var_15806_pad_type_0, strides = var_15802, 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_15806_cast")];
            tensor<fp16, [2, 320, 1, 4096]> inputs_cast = add(x = var_15806_cast, y = inputs_93_cast)[name = tensor<string, []>("inputs_cast")];
            tensor<int32, [1]> var_15810 = const()[name = tensor<string, []>("op_15810"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> channels_mean_cast = reduce_mean(axes = var_15810, keep_dims = var_12802, 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_15814 = const()[name = tensor<string, []>("op_15814"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [2, 1, 1, 4096]> var_15815_cast = reduce_mean(axes = var_15814, keep_dims = var_12802, x = zero_mean_sq_cast)[name = tensor<string, []>("op_15815_cast")];
            tensor<fp16, []> var_15816_to_fp16 = const()[name = tensor<string, []>("op_15816_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [2, 1, 1, 4096]> var_15817_cast = add(x = var_15815_cast, y = var_15816_to_fp16)[name = tensor<string, []>("op_15817_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_15817_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_15821_to_fp16 = const()[name = tensor<string, []>("op_15821_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216760832)))];
            tensor<fp16, [2, 320, 1, 4096]> var_15822_cast = add(x = out_cast, y = var_15821_to_fp16)[name = tensor<string, []>("op_15822_cast")];
            tensor<fp16, [1, 320, 1, 1]> var_15824_to_fp16 = const()[name = tensor<string, []>("op_15824_to_fp16"), val = tensor<fp16, [1, 320, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216761536)))];
            tensor<fp16, [2, 320, 1, 4096]> input_521_cast = mul(x = var_15822_cast, y = var_15824_to_fp16)[name = tensor<string, []>("input_521_cast")];
            tensor<int32, [2]> var_15832 = const()[name = tensor<string, []>("op_15832"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15834 = const()[name = tensor<string, []>("op_15834"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15836_pad_type_0 = const()[name = tensor<string, []>("op_15836_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15836_pad_0 = const()[name = tensor<string, []>("op_15836_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, [204800]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216762240))), lut = tensor<fp16, [4]>([-0x1.814p-4, -0x1.b5p-6, 0x1.dep-6, 0x1.8e4p-4]), 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, [640]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216967104))), lut = tensor<fp16, [4]>([-0x1.51p-4, 0x1.ddp-6, -0x1.d58p-7, 0x1.3ecp-4]), 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_15836_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_0_proj_bias_to_fp16_palettized, dilations = var_15834, groups = var_12807, pad = var_15836_pad_0, pad_type = var_15836_pad_type_0, strides = var_15832, 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_15836_cast")];
            tensor<int32, [2]> var_15837_split_sizes_0 = const()[name = tensor<string, []>("op_15837_split_sizes_0"), val = tensor<int32, [2]>([1280, 1280])];
            tensor<int32, []> var_15837_axis_0 = const()[name = tensor<string, []>("op_15837_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [2, 1280, 1, 4096]> var_15837_cast_0, tensor<fp16, [2, 1280, 1, 4096]> var_15837_cast_1 = split(axis = var_15837_axis_0, split_sizes = var_15837_split_sizes_0, x = var_15836_cast)[name = tensor<string, []>("op_15837_cast")];
            tensor<string, []> var_15839_mode_0 = const()[name = tensor<string, []>("op_15839_mode_0"), val = tensor<string, []>("EXACT")];
            tensor<fp16, [2, 1280, 1, 4096]> var_15839_cast = gelu(mode = var_15839_mode_0, x = var_15837_cast_1)[name = tensor<string, []>("op_15839_cast")];
            tensor<fp16, [2, 1280, 1, 4096]> input_523_cast = mul(x = var_15837_cast_0, y = var_15839_cast)[name = tensor<string, []>("input_523_cast")];
            tensor<int32, [2]> var_15843 = const()[name = tensor<string, []>("op_15843"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15845 = const()[name = tensor<string, []>("op_15845"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15847_pad_type_0 = const()[name = tensor<string, []>("op_15847_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15847_pad_0 = const()[name = tensor<string, []>("op_15847_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, [102400]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216967808))), lut = tensor<fp16, [4]>([-0x1.bd4p-4, -0x1.02p-5, 0x1.044p-5, 0x1.bdcp-4]), 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, []>(217070272)))];
            tensor<fp16, [2, 320, 1, 4096]> var_15847_cast = conv(bias = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_bias_to_fp16, dilations = var_15845, groups = var_12807, pad = var_15847_pad_0, pad_type = var_15847_pad_type_0, strides = var_15843, weight = up_blocks_3_attentions_2_transformer_blocks_0_ff_net_2_weight_to_fp16_palettized, x = input_523_cast)[name = tensor<string, []>("op_15847_cast")];
            tensor<fp16, [2, 320, 1, 4096]> hidden_states_333_cast = add(x = var_15847_cast, y = inputs_cast)[name = tensor<string, []>("hidden_states_333_cast")];
            tensor<int32, [4]> var_15849 = const()[name = tensor<string, []>("op_15849"), val = tensor<int32, [4]>([2, 320, 64, 64])];
            tensor<fp16, [2, 320, 64, 64]> input_525_cast = reshape(shape = var_15849, x = hidden_states_333_cast)[name = tensor<string, []>("input_525_cast")];
            tensor<int32, [2]> var_15853 = const()[name = tensor<string, []>("op_15853"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15855 = const()[name = tensor<string, []>("op_15855"), 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, [25600]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217070976))), lut = tensor<fp16, [4]>([-0x1.0e4p-3, -0x1.48p-5, 0x1.134p-5, 0x1.fdcp-4]), 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, []>(217096640)))];
            tensor<fp16, [2, 320, 64, 64]> hidden_states_cast = conv(bias = up_blocks_3_attentions_2_proj_out_bias_to_fp16, dilations = var_15855, groups = var_12807, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = var_15853, 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, []>(217097344)))];
            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, []>(217098048)))];
            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_15869 = const()[name = tensor<string, []>("op_15869"), val = tensor<int32, []>(1)];
            tensor<int32, [2]> var_15872 = const()[name = tensor<string, []>("op_15872"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_15874 = const()[name = tensor<string, []>("op_15874"), val = tensor<int32, [2]>([1, 1])];
            tensor<string, []> var_15876_pad_type_0 = const()[name = tensor<string, []>("op_15876_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> var_15876_pad_0 = const()[name = tensor<string, []>("op_15876_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, [2880]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217098752))), lut = tensor<fp16, [4]>([-0x1.7c8p-4, -0x1.554p-7, 0x1.8d4p-6, 0x1.be8p-4]), 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]>([-0x1.39p-9, 0x1.9f8p-9, 0x1.c64p-12, 0x1.d6cp-10])];
            tensor<fp16, [2, 4, 64, 64]> var_15876_cast = conv(bias = conv_out_bias_to_fp16, dilations = var_15874, groups = var_15869, pad = var_15876_pad_0, pad_type = var_15876_pad_type_0, strides = var_15872, weight = conv_out_weight_to_fp16_palettized, x = input_cast)[name = tensor<string, []>("op_15876_cast")];
            tensor<string, []> var_15876_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_15876_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [2, 4, 64, 64]> noise_pred = cast(dtype = var_15876_cast_to_fp32_dtype_0, x = var_15876_cast)[name = tensor<string, []>("cast_0")];
        } -> (noise_pred);
}